Window NT vs x86 solaris vs SCO?
Jianlin Chang
More and more sites are deploying Win NT. Besides easy integration with
Windows desktop, low cost is often cited as one major reason of choosing
NT instead of UNIX. What about UNIX on x86 platforms, why aren't they
chosen? I suppose that they shouldn't be much expensive than NT. It
doesn't seem that x86 solaris has gained enough market share, any
reason for this? Thanks.
Edison M. Castro
So, what you are sayng that the UNIX WAY is the one and only right WAY os
doing anything
Wayne Yu
Most people are used to DOS/Windows. NT has that similar face to
most PC users. To most PC users, they feel that UNIX is too complicate
for them. Unfortunately, NT is growing into a similar monster. If you
really know how to run everything on NT, you will feel Solaris for x86
and SCO UNIX too easy. This is easy to understand, for any advanced
OS,
you must to have certain functionalities. NT needs to work harder than
UNIX to achieve the same as UNIX, since it does not want to follow the
UNIX way, which had been improved for many years.
Nils Myklebust
chang@mjs1 (Jianlin Chang) wrote:
:More and more sites are deploying Win NT. Besides easy integration with
:Windows desktop, low cost is often cited as one major reason of choosing
:NT instead of UNIX. What about UNIX on x86 platforms, why aren't they
:chosen? I suppose that they shouldn't be much expensive than NT. It
:doesn't seem that x86 solaris has gained enough market share, any
:reason for this? Thanks.
One major expence of Unix on PCs is to find someone who know anything
about it at all. There are people at every corner that knows Windows
and something about NT.
And have you ever worked with NT?
If you need to install a new program you get a file, click on its
name, do some simple selections in a dialog box and all is fine.
You can start working.
There are no pletora of SSL's or whatever SCO calls them, that you can
hardly find info on. On NT there are service packs. Get the latest
one. Install it like above. It works. Every new service pack includes
everything so you do not need to hunt for a number of things to
install, and install in the right order, and deinstall what you have
installed before and remember to apply or not apply and and and...
You still have drivers for different hardware, but normally they are
relatively easy to install. And all kinds of configurations are mostly
understandable without hunting down a path of manuals.
Unix has significant advantages for many uses. We use it and like it.
But the above is only a small part of the things SCO has to do
something about to even survive in this industry. (They are doing a
little, but is it fast enough. OSE 5 still use shell scripts a lot.)
Solaris (x86) is a more modern unix, but they haven't been able to
market it properly (luckily so we didn't get yet another stupid
variant of Unix in the PC part of the bussiness).
NT is on a rapid path to become a very serious contender, even as a
server for big databases and other complex tasks.
SCO and UnixWare beeing combined to one and the cooperation between HP
and SCO is of course more than a good idea in this situation. It's a
question of survival.
Of course it should have included Sun as well, but the Unix bussiness
managers seems still to be too narrow minded to understand that. It
may be to late when they do. (When HP buy equity in SunSoft (Unix part
only) and SCO is merged with them to form *the* Unix OS company.)
They still believe they have a competitive advantage for making a
"better" Unix. While every user wants one standard Unix, and the
market has been crying out loud for it for years.
Strange. Very strange.
Nils.My...@ccmail.telemax.no
NM Data AS, P.O.Box 9090 Gronland, N-0133 Oslo, Norway
My opinions are those of my company
Stephen Davies
Edison M. Castro wrote:
>
> So, what you are sayng that the UNIX WAY is the one and only right WAY os
> doing anything
I think that sound right!
Evan Leibovitch
In article <4ucul0$q...@knot.queensu.ca>,
Jianlin Chang <ch...@mjs1.phy.queensu.ca> wrote:
>More and more sites are deploying Win NT. Besides easy integration with
>Windows desktop, low cost is often cited as one major reason of choosing
>NT instead of UNIX.
>What about UNIX on x86 platforms, why aren't they chosen?
You've made a few logical leaps here that don't connect. Let's look at
them:
1) Increase of NT's deployment
More sites are deploying NT -- no big surprise, since it is still a
fairly new product. Many, many sites have received evaluation copies,
some of which have led to deployment, many have not.
2) Cost
"Low cost is often cited": by whom? Anyone worth their salaray knows
that the cost of the OS -- *any* OS -- is but a minor part of an
installation's cost. Are NT sysadmins cheaper than Unix ones? How many
CS graduates leave college familiar with NT? How many are familiar with
Unix?
But, OK, let's tackle the issue of cost directly anyway.
How many Unix servers come with per-clients cost to connect? WinNT client
licenses run at $40 per seat. So the low-cost factor isn't really as strong
a point as you suggest. Even the premium-priced SCO bases its user
metrics differently; buying an OS based on simultaneous users, rather
than the number of client stations, can result in a cost savings.
In any event, there's a fairly broad spectrum of Unix prices. If one wants
to be *really* price-sensitive, NT Server will never come close to the
price of any Linux distribution. As low as $10, as high as a whopping $99,
NT-compatible file and printer sharing included, and all the Internet
software one could ask for. In the Linux community it's not a strategic
business move to give away the Web server, it's the normal way of doing
things.
3) NT is chosen instead of Unix
By what poll/survey/study do you come to the fact they're not chosen?
I happen to believe they are; SCO sales are still growing, and the
commercial world is afraid to count Linux installations because they
won't like the results.
It all comes down to a choice of tools for the task. If all that's needed
is merely a file/print server for a bunch of Windows systems, Unix may be
overkill and an NT box may fit the bill just right. SQL Server is an
innovative concept, and it provides a head start for NT in the field of
small databases; but there are already very capable ODBC servers for
Unix that can also be good engines for inexpensive dumb terminals and
Internet use.
>I suppose that they shouldn't be much expensive than NT. It
>doesn't seem that x86 solaris has gained enough market share, any
>reason for this? Thanks.
Solaris makes a good product in Solaris X86, but the company is in a
quandry. Hardware sales are far more lucrative, and if they make
Solaris X86 *too* good it'll cut into SPARC sales. That is something
Sun would dearly want to avoid. SCO and Linux developers, since they
don't sell hardware, have no such artificial restrictions.
--
Evan Leibovitch, Sound Software Ltd., located in beautiful Brampton, Ontario
Caldera Business Partner / SCO Authorized VAR / ev...@telly.org / (905) 452-0504
I love standards because they give non-conformists something not to conform to
John Wiltshire
ev...@telly.org (Evan Leibovitch) wrote:
>2) Cost
>"Low cost is often cited": by whom? Anyone worth their salaray knows
>that the cost of the OS -- *any* OS -- is but a minor part of an
>installation's cost. Are NT sysadmins cheaper than Unix ones? How many
>CS graduates leave college familiar with NT? How many are familiar with
>Unix?
>But, OK, let's tackle the issue of cost directly anyway.
>How many Unix servers come with per-clients cost to connect? WinNT client
>licenses run at $40 per seat. So the low-cost factor isn't really as strong
>a point as you suggest. Even the premium-priced SCO bases its user
>metrics differently; buying an OS based on simultaneous users, rather
>than the number of client stations, can result in a cost savings.
I don't know if you've noticed but NT has the choice of per-client or
per-connection as its licensing model.
>In any event, there's a fairly broad spectrum of Unix prices. If one wants
>to be *really* price-sensitive, NT Server will never come close to the
>price of any Linux distribution. As low as $10, as high as a whopping $99,
>NT-compatible file and printer sharing included, and all the Internet
>software one could ask for. In the Linux community it's not a strategic
>business move to give away the Web server, it's the normal way of doing
>things.
Note you said "all the internet software". There is a lot more
software in the world than Linux software, especially if the company
already has an installed base of PC systems, NT is a fairly good
choice.
>3) NT is chosen instead of Unix
>By what poll/survey/study do you come to the fact they're not chosen?
>I happen to believe they are; SCO sales are still growing, and the
>commercial world is afraid to count Linux installations because they
>won't like the results.
>It all comes down to a choice of tools for the task. If all that's needed
>is merely a file/print server for a bunch of Windows systems, Unix may be
>overkill and an NT box may fit the bill just right.
Novell 4.1 would be a much better solution than NT for simple F&P.
>SQL Server is an
>innovative concept, and it provides a head start for NT in the field of
>small databases; but there are already very capable ODBC servers for
>Unix that can also be good engines for inexpensive dumb terminals and
>Internet use.
NT is good where a high end Unix system would be overkill as a general
rule. Many people are moving towards NT because of its tight
integration with Win 3.x and Win95 - Unix servers don't have this
integration and are unlikely to see the need for it.
For high end apps, NT can't cut it yet. For a cheap upgrade in a
Windows dominated network, NT is a good choice.
>>I suppose that they shouldn't be much expensive than NT. It
>>doesn't seem that x86 solaris has gained enough market share, any
>>reason for this? Thanks.
>Solaris makes a good product in Solaris X86, but the company is in a
>quandry. Hardware sales are far more lucrative, and if they make
>Solaris X86 *too* good it'll cut into SPARC sales. That is something
>Sun would dearly want to avoid. SCO and Linux developers, since they
>don't sell hardware, have no such artificial restrictions.
Sun has done well in its operation and I hope that NT will find its
niche and hopefully MS will develop it to not take over the world but
to fill a need.
Fat chance though...
John Wiltshire
-----------------------------------------------------------
John Wiltshire | Email j...@qits.net.au
Network Systems Engineer | Ph. W +61 7 38342783
Traffic Systems Development | H +61 7 33006052
-----------------------------------------------------------
These opinions may not be mine and are almost certainly
not those of my employer!
Saint (n) -
A dead sinner, revised and edited.
Bryan Althaus
: 3) NT is chosen instead of Unix
:
: By what poll/survey/study do you come to the fact they're not chosen?
: I happen to believe they are; SCO sales are still growing, and the
: commercial world is afraid to count Linux installations because they
: won't like the results.
:
: It all comes down to a choice of tools for the task. If all that's needed
: is merely a file/print server for a bunch of Windows systems, Unix may be
: innovative concept, and it provides a head start for NT in the field of
: small databases; but there are already very capable ODBC servers for
: Unix that can also be good engines for inexpensive dumb terminals and
: Internet use.
AS a PRO UNIX person I will say that NT being chosen over UNIX is real,
and unavoidable simple because NT is the successor to Window95. Most
corporate site are skipping on Window95 and waiting for NT 4.0 which
will come with the Window95 look. True this is the client (NT workStation)
we are talking about and not the server, but once you have NT clients,
the logical choice for most will be an NT server.
Linux will not be hurt by this but SCO surely will. SCO's main business
is in the server area, and NT will hurt SCO's sales, though not nearly
as bad as what it will do to Netware.
For those who have never used NT, it's a nice product mainly because it
was developed by mostly x-Digital top rank and file developers who
Gates pulled in. No way MS could have developed NT on their own. Just
looked at Windows95 for an example of a MS developed OS.
Though, you will notice NT more and more is trying to become like
Windows 95 by brining everything into the kernel for speed, allowing
access to devices that it once didn't, making it faster and more
useable but at the expensive of stability.
Luckily, both the UNIX market is expanding as the NT market is, so it's
a win win situation for all.
:
: >I suppose that they shouldn't be much expensive than NT. It
: >doesn't seem that x86 solaris has gained enough market share, any
: >reason for this? Thanks.
:
: Solaris makes a good product in Solaris X86, but the company is in a
: quandary. Hardware sales are far more lucrative, and if they make
: Solaris X86 *too* good it'll cut into SPARC sales. That is something
: Sun would dearly want to avoid. SCO and Linux developers, since they
: don't sell hardware, have no such artificial restrictions.
:
This is silly. The only thing Sun doesn't do is market Solaris x86,
other than that since it's based on the exact code base as Solaris Sparc
and can't help but get every last feature, CDE 1.2, WABI 2.2, NFS 3.0 etc.
To make things more interesting, now that Sun has decide to dump SBUS
in favor of PCI, what Solaris OS has PCI drivers already for it,
Solaris Intel. In the future both OS's will share PCI device drivers!
This is a quick what's in the upcoming Solaris 2.6, notice how the Intel
version benefits from the Sparc version.
* OEMs can now write source comparable PCMCIA drivers (that will work
everywhere). PCI has the same kind of source compatibility. You
can now take the same driver you write for a x86 PCI card and use it
on a SPARC (news flash: Sun is going towards PCI and leaving SBus
behind).
* SCO binaries will now run on x86
* NFS failover
* Public (web) NFS
* Kernel sockets
* RPC uses doors for really fast local transport.
* CDE is bundled and comes up by default (not unlike HP/UX).
* cachefs improvements that allow use of cache when NFS server is
down.
* Persistent disk IDs (Veritas VM and Solstice DiskSuite will now be happy)
* Spec1170 for UNIX95 branding
* x86 device config. Plug and play is now here!
* X11R6 based X server
Edmond
Bryan Althaus wrote:
> Though, you will notice NT more and more is trying to become like
> Windows 95 by brining everything into the kernel for speed, allowing
> access to devices that it once didn't, making it faster and more
> useable but at the expensive of stability.
Uh, the Win32 subsystem was put into the kernel as were the loading of video
drivers. Is this everything? Only time will tell about stability (no one
has actually tested that part with any authority yet). This doesn't do
didly for device access also. It didn't make it any more usable and it may
have only increased speed marginally for some tasks (very well for others).
--
Edmond Underwood
U Software
Bench32 1.20 - The new threshhold in system benchmarking
Bench32 web site: http://www.rmii.com/~underwoe/bench32.html
Steve Kappel
In article <4une9e$j...@panix.com>, br...@panix.com (Bryan Althaus) writes:
>
>AS a PRO UNIX person I will say that NT being chosen over UNIX is real,
>and unavoidable simple because NT is the successor to Window95. Most
>corporate site are skipping on Window95 and waiting for NT 4.0 which
>will come with the Window95 look. True this is the client (NT workStation)
>we are talking about and not the server, but once you have NT clients,
>the logical choice for most will be an NT server.
>
>Linux will not be hurt by this but SCO surely will. SCO's main business
>is in the server area, and NT will hurt SCO's sales, though not nearly
>as bad as what it will do to Netware.
>
>For those who have never used NT, it's a nice product mainly because it
>was developed by mostly x-Digital top rank and file developers who
>Gates pulled in. No way MS could have developed NT on their own. Just
>looked at Windows95 for an example of a MS developed OS.
It might be a modern product in the kernel but it is still crippled by
a poor API and user environment.
>Though, you will notice NT more and more is trying to become like
>Windows 95 by brining everything into the kernel for speed, allowing
>access to devices that it once didn't, making it faster and more
>useable but at the expensive of stability.
I've read that microkernel architectures, while elegant, don't perform or
scale well. The only major Unix vendor with such a kernel is Digital.
They know they have to do something with NT to fix the scaling problems.
--
______________________________________________________________________
Steve Kappel Work: ste...@apertus.com
Apertus Technologies, Inc. http://www.apertus.com
7275 Flying Cloud Drive Home: ska...@winternet.com
Eden Prairie, MN 55344
Edmond
Steve Kappel wrote:
>
> It might be a modern product in the kernel but it is still crippled by
> a poor API and user environment.
>
poor is certainly a bad word for it. How is it poor? What were you
looking to do that you were so crippled by?
> >Though, you will notice NT more and more is trying to become like
> >Windows 95 by brining everything into the kernel for speed, allowing
> >access to devices that it once didn't, making it faster and more
> >useable but at the expensive of stability.
>
> I've read that microkernel architectures, while elegant, don't perform or
> scale well. The only major Unix vendor with such a kernel is Digital.
> They know they have to do something with NT to fix the scaling problems.
>
Doesn't AIX use a semi-microkernel architecture? I do see your point
about microkernel's and poor scalability. I don't think there are many
scaling problems with NT that deal directly with the microkernel (what
used to be) other than processor scaling. It does run on 32 processor
systems, so it isn't bad. Solaris runs on a 64-CPU box now, so that's
where it stands in comparison. (Yes, I know there are more powerful
machines but rarely are they using commercial UNIXES which was your
comparison).
Jianlin Chang
In many cases, whether a product is successful or not doesn't depend on
whether the product is technologically better. It is pretty certain that NT
will dominate over UNIX in desktop market. You will be surpised that so
many scientific and engineering sites which are previously mainly
UNIX workstation begin to move to NT.
Steve Kappel (ste...@apertus.com) wrote:
: In article <4une9e$j...@panix.com>, br...@panix.com (Bryan Althaus) writes:
: >
: >AS a PRO UNIX person I will say that NT being chosen over UNIX is real,
: >and unavoidable simple because NT is the successor to Window95. Most
: >corporate site are skipping on Window95 and waiting for NT 4.0 which
: >will come with the Window95 look. True this is the client (NT workStation)
: >we are talking about and not the server, but once you have NT clients,
: >the logical choice for most will be an NT server.
: >
: >Linux will not be hurt by this but SCO surely will. SCO's main business
: >is in the server area, and NT will hurt SCO's sales, though not nearly
: >as bad as what it will do to Netware.
: >
: >For those who have never used NT, it's a nice product mainly because it
: >was developed by mostly x-Digital top rank and file developers who
: >Gates pulled in. No way MS could have developed NT on their own. Just
: >looked at Windows95 for an example of a MS developed OS.
: It might be a modern product in the kernel but it is still crippled by
: a poor API and user environment.
: >Though, you will notice NT more and more is trying to become like
: >Windows 95 by brining everything into the kernel for speed, allowing
: >access to devices that it once didn't, making it faster and more
: >useable but at the expensive of stability.
: I've read that microkernel architectures, while elegant, don't perform or
: scale well. The only major Unix vendor with such a kernel is Digital.
: They know they have to do something with NT to fix the scaling problems.
: --
Jianlin Chang
If you look at recent articles at various trade journals, you would know
where what I said come from.
Evan Leibovitch (ev...@telly.org) wrote:
: In article <4ucul0$q...@knot.queensu.ca>,
: Jianlin Chang <ch...@mjs1.phy.queensu.ca> wrote:
: >More and more sites are deploying Win NT. Besides easy integration with
: >Windows desktop, low cost is often cited as one major reason of choosing
: >NT instead of UNIX.
: >What about UNIX on x86 platforms, why aren't they chosen?
: You've made a few logical leaps here that don't connect. Let's look at
: them:
: 1) Increase of NT's deployment
: More sites are deploying NT -- no big surprise, since it is still a
: fairly new product. Many, many sites have received evaluation copies,
: some of which have led to deployment, many have not.
: 2) Cost
: "Low cost is often cited": by whom? Anyone worth their salaray knows
: that the cost of the OS -- *any* OS -- is but a minor part of an
: installation's cost. Are NT sysadmins cheaper than Unix ones? How many
: CS graduates leave college familiar with NT? How many are familiar with
: Unix?
: But, OK, let's tackle the issue of cost directly anyway.
: How many Unix servers come with per-clients cost to connect? WinNT client
: licenses run at $40 per seat. So the low-cost factor isn't really as strong
: a point as you suggest. Even the premium-priced SCO bases its user
: metrics differently; buying an OS based on simultaneous users, rather
: than the number of client stations, can result in a cost savings.
: In any event, there's a fairly broad spectrum of Unix prices. If one wants
: to be *really* price-sensitive, NT Server will never come close to the
: price of any Linux distribution. As low as $10, as high as a whopping $99,
: NT-compatible file and printer sharing included, and all the Internet
: software one could ask for. In the Linux community it's not a strategic
: business move to give away the Web server, it's the normal way of doing
: things.
: 3) NT is chosen instead of Unix
: By what poll/survey/study do you come to the fact they're not chosen?
: I happen to believe they are; SCO sales are still growing, and the
: commercial world is afraid to count Linux installations because they
: won't like the results.
: It all comes down to a choice of tools for the task. If all that's needed
: is merely a file/print server for a bunch of Windows systems, Unix may be
: overkill and an NT box may fit the bill just right. SQL Server is an
: innovative concept, and it provides a head start for NT in the field of
: small databases; but there are already very capable ODBC servers for
: Unix that can also be good engines for inexpensive dumb terminals and
: Internet use.
: >I suppose that they shouldn't be much expensive than NT. It
: >doesn't seem that x86 solaris has gained enough market share, any
: >reason for this? Thanks.
: Solaris makes a good product in Solaris X86, but the company is in a
: quandry. Hardware sales are far more lucrative, and if they make
: Solaris X86 *too* good it'll cut into SPARC sales. That is something
: Sun would dearly want to avoid. SCO and Linux developers, since they
: don't sell hardware, have no such artificial restrictions.
: --
John Wiltshire
ste...@apertus.com (Steve Kappel) wrote:
>In article <4une9e$j...@panix.com>, br...@panix.com (Bryan Althaus) writes:
>>
>>AS a PRO UNIX person I will say that NT being chosen over UNIX is real,
>>and unavoidable simple because NT is the successor to Window95. Most
>>corporate site are skipping on Window95 and waiting for NT 4.0 which
>>will come with the Window95 look. True this is the client (NT workStation)
>>we are talking about and not the server, but once you have NT clients,
>>the logical choice for most will be an NT server.
>>
>>Linux will not be hurt by this but SCO surely will. SCO's main business
>>is in the server area, and NT will hurt SCO's sales, though not nearly
>>as bad as what it will do to Netware.
>>
>>For those who have never used NT, it's a nice product mainly because it
>>was developed by mostly x-Digital top rank and file developers who
>>Gates pulled in. No way MS could have developed NT on their own. Just
>>looked at Windows95 for an example of a MS developed OS.
>It might be a modern product in the kernel but it is still crippled by
>a poor API and user environment.
Why is it a poor API? Seriously, as an NT developer I'd like to know
what I am missing out on.
>>Though, you will notice NT more and more is trying to become like
>>Windows 95 by brining everything into the kernel for speed, allowing
>>access to devices that it once didn't, making it faster and more
>>useable but at the expensive of stability.
>I've read that microkernel architectures, while elegant, don't perform or
>scale well. The only major Unix vendor with such a kernel is Digital.
>They know they have to do something with NT to fix the scaling problems.
NT is slowly growing up to the higher limits of performance. I've
heard that there is currently no point running more than 8 CPUs but it
scales OK up to that (32 CPUs are supported by the kernel at the
moment).
DEC is working on a 64bit addressing system for NT AFAIK to help with
the huge DB support (and to take advantage of their Alphas).
Steve Cole
On 13 Aug 1996, Jianlin Chang wrote:
> In many cases, whether a product is successful or not doesn't depend on
> whether the product is technologically better. It is pretty certain that NT
> will dominate over UNIX in desktop market. You will be surpised that so
> many scientific and engineering sites which are previously mainly
> UNIX workstation begin to move to NT.
I agree, NT is a fairly good desktop and UNIX is largely wasted (in a
potential capacity sense) in that sense. I still love UNIX on the desktop
but NT (4.0 - the 3.51 interface is horrid) also has its place. It's a
damn sight better than most.
Cheers,
Steve |President & Systems Administrator, Kingston Online Services
|613-549-8667 Voice co...@limestone.kosone.com Internet
|(e pluribus unix) 3xT-1! URL: http://www.kosone.com/kos/
|Business and Education partners in SouthEastern Ontario 1993->
|
|"Through the firewall, out the router, down the T1, across the
| backbone, bounced from satellite, it's nothing but net."
Steve Kappel
In article <320FB0...@Colorado.Edu>, Edmond <unde...@Colorado.Edu> writes:
>Steve Kappel wrote:
>>
>
>> It might be a modern product in the kernel but it is still crippled by
>> a poor API and user environment.
>>
>
>looking to do that you were so crippled by?
I'm not a Windows developer but I do develop code that is intended to
be ported to Windows in the future.
Recently I needed to build a generic thread library that I wanted to be
completely portable across several underlying thread APIs (DCE, POSIX,
UI, and NT threads). Thread-specific storage is one case where the
NT API is poor. First, while all implementations have some limit to the number
of thread-specific keys ("locations" in NT terminology), I don't know
of any that are as low as NT which is 64. Second, all the other thread APIs take
a destructor function argument when they create keys (pthread_key_create,
thr_keycreate); this destructor is called when a thread exits passing
the value of the thread-specific data for that key in that thread to it.
This is a nice clean way to prevent memory leaks. The NT API leaves
the programmer to solve this problem - it may seem simple but it
isn't. Is there any way you can reliably be called in NT when a thread
is exiting? In both programs and DLLs? Nope. The closest provided
feature is DLL_PROCESS_DETACH (which itself is a strange concept
compared to Unix shared libraries).
>> >Though, you will notice NT more and more is trying to become like
>> >Windows 95 by brining everything into the kernel for speed, allowing
>> >access to devices that it once didn't, making it faster and more
>> >useable but at the expensive of stability.
>>
>> I've read that microkernel architectures, while elegant, don't perform or
>> scale well. The only major Unix vendor with such a kernel is Digital.
>> They know they have to do something with NT to fix the scaling problems.
>>
>
>Doesn't AIX use a semi-microkernel architecture? I do see your point
>about microkernel's and poor scalability. I don't think there are many
>scaling problems with NT that deal directly with the microkernel (what
>used to be) other than processor scaling. It does run on 32 processor
>systems, so it isn't bad. Solaris runs on a 64-CPU box now, so that's
>where it stands in comparison. (Yes, I know there are more powerful
>machines but rarely are they using commercial UNIXES which was your
>comparison).
The number of processors "supported" and the number that can be
effectively used are two entirely different things. Most people say 4
is about the end of the road for NT, others say 8 max. The efficiency
of NT drops off quickly as you add processors. This is a well-known
phenomena. Solaris scales near linear - just take a look at the SPECrate
numbers for a Cray (now Sun) SuperServer going from 24 to 48 processors
(41967 to 82522 - a ratio of 1.97).
Edmond
John Wiltshire wrote:
>
> ste...@apertus.com (Steve Kappel) wrote:
>
> NT is slowly growing up to the higher limits of performance. I've
> heard that there is currently no point running more than 8 CPUs but it
> scales OK up to that (32 CPUs are supported by the kernel at the
> moment).
>
PC Week said this with no definitive test (Sequent says they never tested their 32
CPU systems). What gives with their comment? Just wondering
--
Edmond Underwood
email: unde...@Colorado.Edu
Bench32 for Windows NT and Windows 95
http://www.rmii.com/~underwoe/bench32.html
Edmond
Steve Kappel wrote:
>
> I'm not a Windows developer but I do develop code that is intended to
> be ported to Windows in the future.
>
> Recently I needed to build a generic thread library that I wanted to be
> completely portable across several underlying thread APIs (DCE, POSIX,
> UI, and NT threads). Thread-specific storage is one case where the
> NT API is poor. First, while all implementations have some limit to the number
> of thread-specific keys ("locations" in NT terminology), I don't know
> of any that are as low as NT which is 64. Second, all the other thread >APIs take
> a destructor function argument when they create keys (pthread_key_create,
> thr_keycreate); this destructor is called when a thread exits passing
> the value of the thread-specific data for that key in that thread to it.
> This is a nice clean way to prevent memory leaks. The NT API leaves
> the programmer to solve this problem - it may seem simple but it
> isn't. Is there any way you can reliably be called in NT when a thread
> is exiting? In both programs and DLLs? Nope. The closest provided
> feature is DLL_PROCESS_DETACH (which itself is a strange concept
> compared to Unix shared libraries).
>
Create a universal thread procedure that waits on executing threads and
signals back to the program when they've finished. Otherwise, use events.
The only fallback I see is the universality of it (what your doing almost
surely be done, but in a different manner under NT and thus not portable).
I don't know about the C Runtime versions which are similar to standard
POSIX format (though not the 0.4 standard in and of itself). Do you
honestly think programming widgets is easier than Windows? Just curious.
> The number of processors "supported" and the number that can be
> effectively used are two entirely different things. Most people say 4
> is about the end of the road for NT, others say 8 max. The efficiency
> of NT drops off quickly as you add processors. This is a well-known
> phenomena. Solaris scales near linear - just take a look at the SPECrate
> numbers for a Cray (now Sun) SuperServer going from 24 to 48 processors
> (41967 to 82522 - a ratio of 1.97).
>
I'm sorry Steve, but do you want to throw a benchmark my way? How poor is
the scaling of NT on 8+ processor systems? How did you test this? Nothing
scales nearly linearly (except in some odd ideal case). There's just too
much overhead involved in coherency, heap allocation from different threads,
etc. Even if it was a shared heap, I think perfect scaling sounds too good
to be true. This looks much like Digital's benchmark showing how their
64-bit CPU makes databses 8x faster than 32-bit CPU's (the data was well
over 6 GB large). There is also no one UNIX. BSDi, NeXTStep, etc. don't
scale at all. Solaris also doesn't scale linearly at all on an x86 dual
P100.
Bryan Althaus
Edmond (unde...@Colorado.Edu) wrote:
It's been said over and over and even MS admit that after 4 CPU's,
NT does not scale well and they are working on it. It just happens
that 4 CPU's is the limit for Intel CPU's which is what 99% of NT
system run on. Digital is a minor player and in two years when
its chapter 11 and bought by Sun no one will remember the company.
Kevin Burton
> I'm sorry Steve, but do you want to throw a benchmark my way? How poor is
> the scaling of NT on 8+ processor systems? How did you test this?
(snip)
Check out http://www.microsoft.com. It is right there. NT 4.0 drops
right off after 8 and 3.51 drops off after 4. Yes, I too believe that
it is impossible to scale across multiple processors linearly however it
all maters what the curve looks like.
Kevin
(snip)
--
(This message sent from within linux slackware 3.0)
Kevin Burton ( kbu...@umbc.edu )
"It is error alone which needs support of the
government. Truth can stand by itself."
--Thomas Jefferson
--
Odd business, marketing. You can take a mediocre
item, spout nonsense, even stuff that's literally
false, and make millions doing it!
--
Bob Palowoda
Edmond (unde...@Colorado.Edu) wrote:
: > NT is slowly growing up to the higher limits of performance. I've
: > heard that there is currently no point running more than 8 CPUs but it
: > scales OK up to that (32 CPUs are supported by the kernel at the
: > moment).
: >
: PC Week said this with no definitive test (Sequent says they never tested their 32
: CPU systems). What gives with their comment? Just wondering
I was wondering what organization such as SPEC (spec rate) uses NT
to make measurements. I noticed that AIM multicpu/users tests
are done on Unix OS not NT. The only benchmark that I have
seen run on NT compared to Unix is TPC. For some reason
I don't see MS making big noise about how their NT runs on
a $975,000 configured box. Not only that you got PCweek
which is geared toward the average joe running a desktop now
how much crediblity and information are you going to get out
of that. If you have that much money to invest your reading the
wrong magazine. That or Sequent couldn't hire enough monkies
to bang on Word and Excel?
---Bob
--
+--------------------------------------------------------+
| palo...@fiver.sns.com http://fiver.sns.com/~palowoda/ |
| Solaris x86 Corner http://fiver.sns.com/ |
+--------------------------------------------------------+
Jouko Holopainen
John Wiltshire wrote:
> >It might be a modern product in the kernel but it is still crippled by
> >a poor API and user environment.
>
> what I am missing out on.
POSIX. You either get native win API (with GUI etc.) OR very limited
subset of posixes.
You cannot use both, because posix is separate subsystem, it is
not atop of NT API.
> NT is slowly growing up to the higher limits of performance. I've
> heard that there is currently no point running more than 8 CPUs but it
> scales OK up to that (32 CPUs are supported by the kernel at the
> moment).
So why it is not a multiuser OS?
This just means that I cannot compile (or access anything
big) over modem.
Please, don't tell me about 3rd party products, this
really should be in the kernel itself.
--
Jouko Holopainen X-Net Oy
Teknologiantie 4 FIN-90570 Oulu
FINLAND GSM: +358 50 5126166
jouko.ho...@xnet.otm.fi
Shawn D. McPeek
br...@panix.com (Bryan Althaus) writes:
>It's been said over and over and even MS admit that after 4 CPU's,
>NT does not scale well and they are working on it. It just happens
>that 4 CPU's is the limit for Intel CPU's which is what 99% of NT
>system run on. Digital is a minor player and in two years when
>its chapter 11 and bought by Sun no one will remember the company.
I don't believe you need to worry about Digital going chapter 11, it's
doing quite well.
Steve Kappel
In article <3210FA...@Colorado.Edu>, Edmond <unde...@Colorado.Edu> writes:
>POSIX format (though not the 0.4 standard in and of itself). Do you
>honestly think programming widgets is easier than Windows? Just curious.
As I said earlier, I'm not a Windows programmer. I've worked with Motif/Xt for
a long time.
For trivial stuff I suppose Windows is probably easier. For more complex things,
like those found in commercial-grade products, Motif/Xt is far more powerful
and any additional complexity resulting from that is minor.
>> The number of processors "supported" and the number that can be
>> effectively used are two entirely different things. Most people say 4
>> is about the end of the road for NT, others say 8 max. The efficiency
>> of NT drops off quickly as you add processors. This is a well-known
>> phenomena. Solaris scales near linear - just take a look at the SPECrate
>> numbers for a Cray (now Sun) SuperServer going from 24 to 48 processors
>> (41967 to 82522 - a ratio of 1.97).
>>
>
>I'm sorry Steve, but do you want to throw a benchmark my way? How poor is
>the scaling of NT on 8+ processor systems? How did you test this? Nothing
Another post says that the MS Web site admits that 3.51 drops off after 4
and 4.0 after 8.
>scales nearly linearly (except in some odd ideal case). There's just too
>much overhead involved in coherency, heap allocation from different threads,
>etc. Even if it was a shared heap, I think perfect scaling sounds too good
>to be true. This looks much like Digital's benchmark showing how their
>64-bit CPU makes databses 8x faster than 32-bit CPU's (the data was well
>over 6 GB large). There is also no one UNIX. BSDi, NeXTStep, etc. don't
I'd say those SPECrate numbers are pretty damn close to linear. Notice
I didn't put out numbers for going from 1 to 2 or 2 to 3 but 24 to 48. Any
significant degradation should easily show up at that scale.
Do you understand what makes up the SPEC benchmarks? Take a look
at http://www.specbench.org/osg . These are every-day real-world
things. No huge address space tricks.
Bryan Althaus
Shawn D. McPeek (smc...@coewl.cen.uiuc.edu) wrote:
Hello, they just lost ~$300 million last quarter. It's the first lose
in about 2 years. There PC division is doing badly, and they have
no quick fix yet.
Just to let you see what 300 million $$$ looks like,
- $300,000,000.00
Evan Leibovitch
In article <4uondm$f...@knot.queensu.ca>,
Jianlin Chang <ch...@mjs1.phy.queensu.ca> wrote:
>If you look at recent articles at various trade journals, you would know
>where what I said come from.
Absolutely. Check out:
PC Week, August 4, 1996: "Unix: The Antidote for NT Overload" by Eric Lundquist
In the same issue of PC Week, reviewing NT 4.0:
"NT is a serious server OS for businesses, but there are other OSes (eg
NetWare, Unix and MVS) that are equal if not better"
Some of your "various trade journals" must include Microsoft
cheerleaders such as Byte, which predicted Unix would be dead and
forgotten by last year. Some of the more balanced media such as
Information Week isn't quite so glazed-over.
Edmond
Kevin Burton <kbu...@umbc.edu> wrote:
>> I'm sorry Steve, but do you want to throw a benchmark my way? How poor is
>> the scaling of NT on 8+ processor systems? How did you test this?
>
>
>Check out http://www.microsoft.com. It is right there. NT 4.0 drops
>right off after 8 and 3.51 drops off after 4. Yes, I too believe that
>it is impossible to scale across multiple processors linearly however it
>all maters what the curve looks like.
>
I didn't see that (show the actual URL as nothing was RIGHT THERE). The
standard specifications drop off after 8 (Corollary's C-BUS II
architecture) and after 4 (MPS 1.4).
-----------------------------------------------------------------------
Edmond Underwood
U Software
E-mail: unde...@rmii.com
Bench32 1.20 - The new threshhold in system benchmarking.
Bench32 web site: http://www.rmii.com/~underwoe/bench32.html.
Edmond
Edmond
ste...@apertus.com (Steve Kappel) wrote:
>As I said earlier, I'm not a Windows programmer. I've worked with Motif/Xt for
>a long time.
>
>For trivial stuff I suppose Windows is probably easier. For more complex things,
>like those found in commercial-grade products, Motif/Xt is far more powerful
>and any additional complexity resulting from that is minor.
>
Both are pretty powerful.
>>I'm sorry Steve, but do you want to throw a benchmark my way? How poor is
>>the scaling of NT on 8+ processor systems? How did you test this? Nothing
>
>Another post says that the MS Web site admits that 3.51 drops off after 4
>and 4.0 after 8.
>
This is a major cop out. You should have spoken up. It looks like you don't
have much. Again, the SPECs drop off at 4 and 8, but not the OS. If it did, it
would be put on systems with 16 and 32 CPU's (why would someone put a 4 CPU OS
on a 32 CPU system). Again, no official results. I'm looking for something much
better than a cop out.
>I'd say those SPECrate numbers are pretty damn close to linear. Notice
>I didn't put out numbers for going from 1 to 2 or 2 to 3 but 24 to 48. Any
>significant degradation should easily show up at that scale.
>
>Do you understand what makes up the SPEC benchmarks? Take a look
>at http://www.specbench.org/osg . These are every-day real-world
>things. No huge address space tricks.
>
I was making a comparison. SPEC isn't even a parallel benchmark. The folks who
made that test did some strange parallelisms to SPEC to get it to scale to 48
CPU's. Who knows, the results could have been placed on threads where the return
counters didn't actually wait for total thread completion. Maybe it's one of
those compiler's that looks for parallelism's in code (look at last month's UNIX
Review of Sun's compiler) and pretty much does the prior (not to mention changes
the way the algorithm works). I'm not buying the result either which way.
Anyway, I've seen Solaris on x86 dual P100's where it doesn't come close to that.
That pretty much defeats the Solaris scales nearly perfectly line. Seeing is
believing is knowing.
Edmond
br...@panix.com (Bryan Althaus) wrote:
>It's been said over and over and even MS admit that after 4 CPU's,
>NT does not scale well and they are working on it. It just happens
>that 4 CPU's is the limit for Intel CPU's which is what 99% of NT
>system run on. Digital is a minor player and in two years when
>its chapter 11 and bought by Sun no one will remember the company.
Show me WHERE EXACTLY SOMEONE TESTED more than 4 CPU's and SHOWED THAT NT
FELL OFF DRASTICALLY!!!!! Until you come up with more than that it's been
said crap, I'm just not buying it. I've seen results of 8 CPU systems, so
perhaps I'm not quite as easy to convince.
Wayne Hyde
Bryan Althaus <br...@panix.com> wrote in article <4ut4ha$5...@panix.com>...
And IBM lost over 6 _BILLION_ a few years back. DEC is a big company.
They will survive a 300M loss.
--
Wayne Hyde | System Administrator
w...@cis.ufl.edu | Fla. Cooperative Fish & Wildlife Research
Unit
http://www.cis.ufl.edu/~wjh | and the
I speak for me, not them -> | U of Florida Department of Surveying and
Mapping
Bryan Althaus
Edmond (unde...@rmii.com) wrote:
: br...@panix.com (Bryan Althaus) wrote:
:
: >It's been said over and over and even MS admit that after 4 CPU's,
: >NT does not scale well and they are working on it. It just happens
: >that 4 CPU's is the limit for Intel CPU's which is what 99% of NT
: >system run on. Digital is a minor player and in two years when
: >its chapter 11 and bought by Sun no one will remember the company.
:
: Show me WHERE EXACTLY SOMEONE TESTED more than 4 CPU's and SHOWED THAT NT
: FELL OFF DRASTICALLY!!!!! Until you come up with more than that it's been
: said crap, I'm just not buying it. I've seen results of 8 CPU systems, so
: perhaps I'm not quite as easy to convince.
First don't scream. Second unless you live under a rock, every major
computer publication has said NT does not scale well beyond 4 CPU's.
I guess there wrong and your right. This was for NT 3.51 and things
will get better.
You can of course run NT and any other SMP OS on more CPU's, but if
they don't scale well, each CPU you add gets you very little after
a certain #. On Solaris it's somewhere between 32-64.
Bryan Althaus
Wayne Hyde (w...@cis.ufl.edu) wrote:
: Bryan Althaus <br...@panix.com> wrote in article <4ut4ha$5...@panix.com>...
: > Shawn D. McPeek (smc...@coewl.cen.uiuc.edu) wrote:
:
: > : I don't believe you need to worry about Digital going chapter 11, it's
: > : doing quite well.
:
: > Hello, they just lost ~$300 million last quarter. It's the first lose
: > in about 2 years. There PC division is doing badly, and they have
: > no quick fix yet.
:
: And IBM lost over 6 _BILLION_ a few years back. DEC is a big company.
: They will survive a 300M loss.
IBM has done a great turn around and has not lost money in any quarter
since including some quarters with HUGE profits. DEC after having
HUGE loses, in the Billions, and having lost alot of market share, who runs
VMS anymore?, had been profitable for the last 7 quarters. It's not the
~$300 million that is going to hurt DEC, it's just unlike IBM, DEC's reorg
is now taking a turn for the worse.
Just look at DEC's stock price (38 today, 52 week high 76.50) to give you
an idea where others think there headed. Other than the Alpha chip, DEC is DOA.
Edmond
ste...@apertus.com (Steve Kappel) wrote:
>
>How convenient! First you say that this is a "cop out" and then you apparently
>agree that SPECs do drop off at 4 and 8. Hmmm.... Of course I wonder where
>you got that - I don't believe that the SPEC suite runs on NT. Study-up on what
NO!!!! The specifications (not SPEC integer or FP or something like that).
>SPECrate is and you will find that it is a good general-purpose multiuser/
>multiprocessor benchmark of real-world usage. Clearly you are more interested
>in getting the answer you want - if not then the benchmark is no good.
>
Your losing it here. I don't think any aircraft made will reach the speed of
light no matter what benchmark said it did. I have reason to doubt such.
[snipped]
>
>>I was making a comparison. SPEC isn't even a parallel benchmark. The folks who
>>made that test did some strange parallelisms to SPEC to get it to scale to 48
>>CPU's. Who knows, the results could have been placed on threads where the return
>>counters didn't actually wait for total thread completion. Maybe it's one of
>>those compiler's that looks for parallelism's in code (look at last month's UNIX
>>Review of Sun's compiler) and pretty much does the prior (not to mention changes
>>the way the algorithm works). I'm not buying the result either which way.
>
>So you think you need a parallel benchmark to see scalability eh? Might make sense
>in the single-user world of NT but Unix is multiuser/multiprocess. Databases
>like Oracle use many processes and it maps really well to a SMP machine *without*
>being multithreaded (apparently what you mean by "parallel"). That is exactly what
Multithreaded doesn't necessarily mean parallel (I can see we're talking a
different language).
>SPECrate measures. The difference between multithreaded and multiprocess
>applications in Unix is simply a matter of efficiency. SPECrate is heavily oriented
>towards multiprocess - and any multithread-oriented benchmark is going to have
>better results.
>
>Again, not the results you want so the benchmark is no good.
>
I'm questioning the accuracy and you totally avoided the question. Why don't
you make another run of that. I made clear and concise points about validity
of such and you ignored it. I'm saying that your benchmark doesn't show
scalability of the OS. If you don't believe me, write to Jeff Reily of SPECorg
and ask him yourself. Just because the benchmark is parallel, doesn't mean
that your showing scalability.
>>Anyway, I've seen Solaris on x86 dual P100's where it doesn't come close to that.
>>That pretty much defeats the Solaris scales nearly perfectly line. Seeing is
>>believing is knowing.
>
>Which version? Solaris x86 is relatively new and only recently has been well-
>tuned for x86. The SPEC data is for SPARC, not x86. It may even be that
>Intel machines are inherently incapable of scaling in SMP environments.?
>What is the NT scalability on Alpha's? Or are we back to Intel being the only
>relevant processor?
I don't have any data on the Alpha. Sorry, I really should but I don't.
Solaris 2.5 (haven't tried 2.5.1). Doesn't matter though, we have SS20's with
dual modules in them and they aren't scaling perfectly. I do agree that
hardware is a key issue.
Steve Cole
On 12 Aug 1996, Bryan Althaus wrote:
> * CDE is bundled and comes up by default (not unlike HP/UX).
CDE has been bundled since v2.5, but I have been assured by my Sun rep
that CDE is an optional install and will not be a default requirement.
This is good news, AFAIC.
You forgot further Ultra architecture improvements and 64-bitness where it
makes sense to have it.
Edmond
br...@panix.com (Bryan Althaus) wrote:
>Edmond (unde...@rmii.com) wrote:
>: br...@panix.com (Bryan Althaus) wrote:
>:
>: >It's been said over and over and even MS admit that after 4 CPU's,
>: >NT does not scale well and they are working on it. It just happens
>: >that 4 CPU's is the limit for Intel CPU's which is what 99% of NT
>: >system run on. Digital is a minor player and in two years when
>: >its chapter 11 and bought by Sun no one will remember the company.
>:
>: Show me WHERE EXACTLY SOMEONE TESTED more than 4 CPU's and SHOWED THAT NT
>: FELL OFF DRASTICALLY!!!!! Until you come up with more than that it's been
>: said crap, I'm just not buying it. I've seen results of 8 CPU systems, so
>: perhaps I'm not quite as easy to convince.
>:
>First don't scream. Second unless you live under a rock, every major
>computer publication has said NT does not scale well beyond 4 CPU's.
>I guess there wrong and your right. This was for NT 3.51 and things
>will get better.
Show me (yes I'll scream until your thick skull gets this through it) 1
PUBLICATION that TESTS NT on greater than 4 CPU systems!!! Show me NOW!
Show me the evidence of such. Show me where they said it and how they tested
this. I'm not living under a rock, I've seen the publications. 1 quote from
PC Week quoted a manufacturer of the new 8 CPU systems that said scalability
would drop off after 8 (gosh do you think they said this to sell their
technology). I'm sure you could find a manufacturer who sold a 4 CPU system
that said that it falls off after 4. You could find a 16 CPU manufacturer
that said it falls off after 16 and you could find a 32 CPU manufacturer that
would say it falls off after 32. Which would you find more of? Which had a
definitive test to show this? I'm not asking too much at all when I say you
need some form of proof. This is as bad as the OS/2 guy who says that OS/2
scales up to 64 processors even though there isn't 1 64 processor systems
that it runs on to test such a theory. That's what you call a junk argument.
Just show me some data and I'll be somewhat appeased.
>You can of course run NT and any other SMP OS on more CPU's, but if
>they don't scale well, each CPU you add gets you very little after
>a certain #. On Solaris it's somewhere between 32-64.
This is ridiculously wrong. You can't just run any SMP OS on more than the
number of CPU's it's set for. SMP architectures don't grow on trees.
Hardware manufacturers must develop these architectures with specific logic
to get the job done. Linux wouldn't run on a Sequent box (the kernel
wouldn't recognize the architecture) and neither would most other SMP OS's.
Most PC oriented SMP OS's usually adhere to some specification like MPS
(which is limited to now 4 processor systems in version 1.4). An MPS system
can't be made with more than that many processors, thus any SMP OS that is
only MPS compliant won't run on other systems. Corollary has many patents
for higher order systems (the original 4 CPU systems out years ago used
Corollary's C-Bus architecture). If your systems kernel isn't designed for
that architecture, then it won't run on those systems. Solaris was
redevelped for the 64 CPU system. I can guarantee you that those boxes have
a different Solaris kernel. Recall that you address hardware within the
software. If it changes, you must make those changes. How do you think
Sequent got NT running on greater than 8 CPU systems? It had to develop it's
own abstraction layer (and also changed the kernel a little to address bus
issues). SUN makes their own hardware and thus can taylor their OS to their
hardware. Let's see how Solaris runs on one of those Suquent boxes (if it
will run at all). It just so happens that NT 4.0 was taylored to run on
those 8 CPU systems by MS itself. MS likely believes the architecture was
important enough to change NT to favor it. Other manufacturers must do this
on their own, thus their "UNPROVEN" as one magazine correctly quoted (PC
Week) on systems with greater than CPU's. Unproven to everyone other than
the manufacturer and their customers of course. Anyway, let's see the tests
that shows it drops off first.
David Eckhardt
Wayne Yu wrote:
>
> In <4ucul0$q...@knot.queensu.ca> chang@mjs1 (Jianlin Chang) writes:
> >
> >More and more sites are deploying Win NT. Besides easy integration
> with
> >Windows desktop, low cost is often cited as one major reason of
> choosing
> >NT instead of UNIX. What about UNIX on x86 platforms, why aren't they
> >doesn't seem that x86 solaris has gained enough market share, any
> >reason for this? Thanks.
>
> most PC users. To most PC users, they feel that UNIX is too complicate
> for them. Unfortunately, NT is growing into a similar monster. If you
> really know how to run everything on NT, you will feel Solaris for x86
> and SCO UNIX too easy. This is easy to understand, for any advanced
> OS,
> you must to have certain functionalities. NT needs to work harder than
> UNIX to achieve the same as UNIX, since it does not want to follow the
> UNIX way, which had been improved for many years.
There are many things that UNIX can do better, but there are many things
that NT can do better. In a smaller orgaization (5 person) there is no
need for UNIX unless you need a DNS for internet travel.
Larger corporations should and are using a mix of NT and UNIX. NT is a
very good operating system for housing Applications that run on the
Microsoft systems.
If your users are using PC's with Win95 and/or Win 3.11 then it would be
easier to setup a Microsoft Network using NT instead of UNIX. If you
use UNIX you will have a protocal barrier that will need NFS to cross.
I am not saying to quit using UNIX because UNIX does have its place, but
NT is a very good and easy to use system if you are dealing with
computers with a Microsoft OS.
I have not spent alot of time with UNIX because my school is teaching
non UNIX based Netware Networks. This gives me a disadvantage with
dealing with UNIX, but I have just dived into Solaris x86 2.5 and found
it interesting, but different then what I am use too.
NT only took me 1 Month to understand and beable to use. Because UNIX is
not in the same format that I am use too then I will have to learn that
before I can get into Network administering. Most students see
Microsoft all the time before atending a UNIX class. That makes it
easier for a student to decide on NT instead of UNIX, but that could be
a mistake.
UNIX and NT are both very good and strong operating systems.
David Eckhardt
TI Stafford Network Support
d-eck...@ti.com
Lucky Leavell
On Thu, 15 Aug 1996, David Eckhardt wrote:
> If your users are using PC's with Win95 and/or Win 3.11 then it would be
> easier to setup a Microsoft Network using NT instead of UNIX. If you
> use UNIX you will have a protocal barrier that will need NFS to cross.
>
Not any longer! See SCO's VisionFS announcement last week. It requires no
special client-side software and outperforms NFS. Check it out on
and even download a 30 day demo.
FYI,
Lucky
Lucky Leavell Phone: (812) 945-6555
Relational Information Systems, Inc. FAX: (812) 949-9233
4727 Grant Line Road Email: r...@iglou.com (Internet)
New Albany, IN 47150-2013 71534,2674 (CompuServe)
WWW Home Page: http://www.iglou.com/ris ftp://www.iglou.com/members/ris
Steve Kappel
In article <4utfqg$j...@lace.colorado.edu>, Edmond <unde...@rmii.com> writes:
>ste...@apertus.com (Steve Kappel) wrote:
>
>>>I'm sorry Steve, but do you want to throw a benchmark my way? How poor is
>>>the scaling of NT on 8+ processor systems? How did you test this? Nothing
>>
>>Another post says that the MS Web site admits that 3.51 drops off after 4
>>and 4.0 after 8.
>>
>
>This is a major cop out. You should have spoken up. It looks like you don't
>have much. Again, the SPECs drop off at 4 and 8, but not the OS. If it did, it
>would be put on systems with 16 and 32 CPU's (why would someone put a 4 CPU OS
>on a 32 CPU system). Again, no official results. I'm looking for something much
>better than a cop out.
How convenient! First you say that this is a "cop out" and then you apparently
agree that SPECs do drop off at 4 and 8. Hmmm.... Of course I wonder where
you got that - I don't believe that the SPEC suite runs on NT. Study-up on what
SPECrate is and you will find that it is a good general-purpose multiuser/
multiprocessor benchmark of real-world usage. Clearly you are more interested
in getting the answer you want - if not then the benchmark is no good.
I'd love to hear specifics of sites running >8 CPU NT machines. Machines running
internal to the hardware vendor don't count; give me end-user production systems.
I certainly read alot about Unix servers doing it and know of several right here
in town.
Also remember that just because a site is running such a configuration it
doesn't mean that it performs well or is cost-effective. I can't name vendors
but we have an 8 processor Unix-based monster here that is a dog. No doubt
a Unix implementation can be just as bad as NT. But then again this thread
is about Solaris, isn't it...
>>I'd say those SPECrate numbers are pretty damn close to linear. Notice
>>I didn't put out numbers for going from 1 to 2 or 2 to 3 but 24 to 48. Any
>>significant degradation should easily show up at that scale.
>>
>>Do you understand what makes up the SPEC benchmarks? Take a look
>>at http://www.specbench.org/osg . These are every-day real-world
>>things. No huge address space tricks.
>>
>
>I was making a comparison. SPEC isn't even a parallel benchmark. The folks who
>made that test did some strange parallelisms to SPEC to get it to scale to 48
>CPU's. Who knows, the results could have been placed on threads where the return
>counters didn't actually wait for total thread completion. Maybe it's one of
>those compiler's that looks for parallelism's in code (look at last month's UNIX
>Review of Sun's compiler) and pretty much does the prior (not to mention changes
>the way the algorithm works). I'm not buying the result either which way.
So you think you need a parallel benchmark to see scalability eh? Might make sense
in the single-user world of NT but Unix is multiuser/multiprocess. Databases
like Oracle use many processes and it maps really well to a SMP machine *without*
being multithreaded (apparently what you mean by "parallel"). That is exactly what
SPECrate measures. The difference between multithreaded and multiprocess
applications in Unix is simply a matter of efficiency. SPECrate is heavily oriented
towards multiprocess - and any multithread-oriented benchmark is going to have
better results.
Again, not the results you want so the benchmark is no good.
>Anyway, I've seen Solaris on x86 dual P100's where it doesn't come close to that.
Steve Kappel
In article <4uve9s$r...@lace.colorado.edu>, Edmond <unde...@rmii.com> writes:
>ste...@apertus.com (Steve Kappel) wrote:
>
>>SPECrate is and you will find that it is a good general-purpose multiuser/
>>multiprocessor benchmark of real-world usage. Clearly you are more interested
>>in getting the answer you want - if not then the benchmark is no good.
>
>light no matter what benchmark said it did. I have reason to doubt such.
Like it or not, SPECrate for a 24 processor Solaris/SPARC machine is 41967 and
a 48 processor is 82522 - a ratio of 1.97 (with 2.0 being perfect).
You can bitch about what SPECrate is all you want but it is a "good" benchmark.
None are perfect. The fact is that in large configurations the ratio is almost
linear.
You complain about not seeing results that *prove* NT doesn't scale well. Yet
you are unable to produce results that prove NT *does* scale well. If so many
people are doing this and the results are so good, why the hell don't they
publish it all over the place? My common sense says there is a reason why
they don't publish it. I've given you a solid benchmark result that would cause
any rational person to conclude that Solaris/SPARC scales _very_ well. Show
me any kind of broad NT benchmark that supports your opinion. How about
Oracle TPC results?
In article <4uvfd7$r...@lace.colorado.edu>, Edmond <unde...@rmii.com> writes:
>Show me (yes I'll scream until your thick skull gets this through it) 1
>PUBLICATION that TESTS NT on greater than 4 CPU systems!!! Show me NOW!
>Show me the evidence of such. Show me where they said it and how they tested
>this. I'm not living under a rock, I've seen the publications. 1 quote from
>PC Week quoted a manufacturer of the new 8 CPU systems that said scalability
>would drop off after 8 (gosh do you think they said this to sell their
>technology). I'm sure you could find a manufacturer who sold a 4 CPU system
>that said that it falls off after 4. You could find a 16 CPU manufacturer
>that said it falls off after 16 and you could find a 32 CPU manufacturer that
>would say it falls off after 32. Which would you find more of? Which had a
>definitive test to show this? I'm not asking too much at all when I say you
I wouldn't rely on a "Publication" for benchmarking large-scale
systems. They simply wouldn't have a clue how to do it. "PC Week"? Now
there's an experienced large-systems bunch!
I would be the first to agree that SPEC or somebody needs a standard
benchmark that is portable to Unix and NT for apples-to-apples comparison.
But O/S scalability is something that can be estimated with any good
system-intensive benchmark. You don't need to compare NT vs. Solaris for
a given # of CPUs and CPU speed doesn't matter. You draw a graph of the
benchmark with points for 1, 2, 4, 8, 16, 32, 64, ... processors. Do that for
Solaris/SPARC/SPECrate and you will see a nice nearly-linear graph. Do that
for NT/Intel/anything-you-can-find and you will see a very limp arrow.
>need some form of proof. This is as bad as the OS/2 guy who says that OS/2
>scales up to 64 processors even though there isn't 1 64 processor systems
>that it runs on to test such a theory. That's what you call a junk argument.
>Just show me some data and I'll be somewhat appeased.
OK, show me the data for 1 32 processor NT system! Come on now, it supports
32, right? Your vapor is no less so than OS/2. Go get the damn SPECrate
table and you will see numbers on Solaris/SPARC ALL THE WAY TO 64 PROCESSORS.
Come on! Digital has big monsters. Sequents run NT? NCR? Come on, where
is the graph from ANY of these vendors? If I were a marketing guy in any
of these companies I'd be splattering that data in Time magazine if it was
good!
Put up or shut up.
Edmond
ste...@apertus.com (Steve Kappel) wrote:
>In article <4uve9s$r...@lace.colorado.edu>, Edmond <unde...@rmii.com> writes:
>>ste...@apertus.com (Steve Kappel) wrote:
>>
>>>SPECrate is and you will find that it is a good general-purpose multiuser/
>>>multiprocessor benchmark of real-world usage. Clearly you are more interested
>>>in getting the answer you want - if not then the benchmark is no good.
>>
>>Your losing it here. I don't think any aircraft made will reach the speed of
>>light no matter what benchmark said it did. I have reason to doubt such.
>
>Like it or not, SPECrate for a 24 processor Solaris/SPARC machine is 41967 and
>a 48 processor is 82522 - a ratio of 1.97 (with 2.0 being perfect).
>
I don't think your understanding the result. SPEC is not internally meant to be a
multi-CPU benchmark. Vendors are using special compilers that are looking for
parallel sequences and actually changing the behavior of the benchmark. You again
ignored the issues I addressed with that. You also have NO PARALLEL to NT running
SPEC with the same compiler. Even if you did, "The results for SPEC wouldn't
change by running it under Windows NT". "SPEC was written in such a way that it's
platform/OS independent", Jeff Reily, from SPEC org from comp.benchmarks. He then
went on to say that you'd only be testing the differences between compilers by
switching from UNIX to NT. SPEC is thus a hardware benchmark. You would have
gotten the same ratio with any other OS on that same hardware platform. I'm not
even sure about the result as it is since the benchmark changes when compiled in
that way.
>You can bitch about what SPECrate is all you want but it is a "good" benchmark.
>None are perfect. The fact is that in large configurations the ratio is almost
>linear.
>
Recall that a year ago, we found out that Intel's SPEC92 results were off. It's
not SPEC's fault, but vendors trying to squeeze the best result out any which way
they can. I don't even think you can compare that result to anything. I'm still
not even sure how it would show that Solaris scales. What goes into OS scaling
(yes here we go again with this one)? Which part of scaling belongs to the OS
(kernel level) and which part to the application (user level)? Which part
determines how well an OS scales? Does the algorithm used have anything to do
with how well something scales? (I know the answer to these, but I think you need
to answer them for yourself).
>You complain about not seeing results that *prove* NT doesn't scale well. Yet
>you are unable to produce results that prove NT *does* scale well. If so many
You need to read the questions I wrote above before we can go on here. Testing
how well the OS ALONE scales is difficult quite honestly. There are so many other
factors involved when benchmarking.
>people are doing this and the results are so good, why the hell don't they
>publish it all over the place? My common sense says there is a reason why
>they don't publish it. I've given you a solid benchmark result that would cause
>any rational person to conclude that Solaris/SPARC scales _very_ well. Show
>me any kind of broad NT benchmark that supports your opinion. How about
>Oracle TPC results?
>
People aren't doing anything except talking at the mouth such as yourself. No one
is proving anything because they don't have the data to. You STILL haven't come
up with the quotes where people are saying these things all over the place. I've
read what they've said in PC Week and InfoWorld and gave my reasons why. These
quotes were from specific manufacturers who produced SMP hardware running NT.
NT's scaling broke off where their specifications did. In each case, they were
different. Is Oracle designed to run on NT systems with more than 8 processors?
I'd like to here from an Oracle representative about that one.
For instance, SQL Server isn't designed to run on systems with more than 6
processors since the other folks who were using it on their machines were at most
using the 6 processor systems from Intergraph. I do know of results for greater
than 8 CPU systems (Hmmm... now how would I get those). They have little to do
with NT's ability and more to do with the hardware though. Their also at the
discretion of the manufacturer and not my own. I wouldn't need them to end this
argument anyway as you have no proof of either.
>>Show me (yes I'll scream until your thick skull gets this through it) 1
>>PUBLICATION that TESTS NT on greater than 4 CPU systems!!! Show me NOW!
>>Show me the evidence of such. Show me where they said it and how they tested
>>this. I'm not living under a rock, I've seen the publications. 1 quote from
>>PC Week quoted a manufacturer of the new 8 CPU systems that said scalability
>>would drop off after 8 (gosh do you think they said this to sell their
>>technology). I'm sure you could find a manufacturer who sold a 4 CPU system
>>that said that it falls off after 4. You could find a 16 CPU manufacturer
>>that said it falls off after 16 and you could find a 32 CPU manufacturer that
>>would say it falls off after 32. Which would you find more of? Which had a
>>definitive test to show this? I'm not asking too much at all when I say you
>
>I wouldn't rely on a "Publication" for benchmarking large-scale
>systems. They simply wouldn't have a clue how to do it. "PC Week"? Now
>there's an experienced large-systems bunch!
>
PC Week simply said that it was unproven. They did no benchmarking. Few have and
you have no results from any. They quoted a manufacturer of Corollary's new C-Bus
II architecture (an 8 CPU architecture) that said that NT wouldn't scale past 8
processors. The manufacturer doesn't even make a 16 processor system nor is there
a specification for one in the x86 world. Some vendors like Sequent, AST (I
believe or it's another acronym that starts with 'A'), have built their own
proprietary architectures and thus had to do the porting themselves. Again, MS
can't address every architecture in NT. You've got to set a target. What scales
well for one may not scale well for another. The proprietary vendors had to
ensure that it scaled on their architectures by making changes themselves.
>I would be the first to agree that SPEC or somebody needs a standard
>benchmark that is portable to Unix and NT for apples-to-apples comparison.
>But O/S scalability is something that can be estimated with any good
>system-intensive benchmark. You don't need to compare NT vs. Solaris for
>a given # of CPUs and CPU speed doesn't matter. You draw a graph of the
>benchmark with points for 1, 2, 4, 8, 16, 32, 64, ... processors. Do that for
>Solaris/SPARC/SPECrate and you will see a nice nearly-linear graph. Do that
>for NT/Intel/anything-you-can-find and you will see a very limp arrow.
>
You really don't understand benchmarking at all. That's my biggest beef with you.
You come off with too much talk and don't understand the underlying issues.
Answer the questions I've drawn out for you before. You may learn much in the
process.
>OK, show me the data for 1 32 processor NT system! Come on now, it supports
>32, right? Your vapor is no less so than OS/2. Go get the damn SPECrate
>table and you will see numbers on Solaris/SPARC ALL THE WAY TO 64 PROCESSORS.
>Come on! Digital has big monsters. Sequents run NT? NCR? Come on, where
>is the graph from ANY of these vendors? If I were a marketing guy in any
>of these companies I'd be splattering that data in Time magazine if it was
>good!
>
Where is the parallel compiler for x86 NT based systems? Where is SPEC for NT?
Where is the 4 CPU Intel systems of any kind? Where is the 2 CPU systems of Intel
systems of any kind? How does SPEC show how well the OS scales when even SPEC
admits that it doesn't. You can jump onto comp.benchmarks any time and tell
everyone that SPEC tests how well the OS scales.
>Put up or shut up.
>
This is what you really need to do.
Edmond
Lucky Leavell <r...@iglou.com> wrote:
>On Thu, 15 Aug 1996, David Eckhardt wrote:
>
>> If your users are using PC's with Win95 and/or Win 3.11 then it would be
>> easier to setup a Microsoft Network using NT instead of UNIX. If you
>> use UNIX you will have a protocal barrier that will need NFS to cross.
>>
>Not any longer! See SCO's VisionFS announcement last week. It requires
What about Samba?
Steve Cole
A note on VisionFS - without crawling over NDA too far, I can say that Sun
may also release a new Wintel to Solaris solution soon.
Steve Cole
I know this is hard for you, Edmond. But shouldn't every compiler which
runs on multi-processing machines actually compile the code as efficiently
for parallel processing as possible? What it does for benchmarks it can
do for other code, right?
In other words, are you saying that NT's only problem is that it doesn't
have a good compiler? Or are you saying that compilers on everything but
NT are so clever that benchmarks mean nothing anyways?
Bryan Althaus
Edmond (unde...@rmii.com) wrote:
: Also, there are in house companies everywhere which will have differing
: results than you. I guess it would depend on what they were doing and
: what systems they were running. Without knowing what they were doing, I
: just can't accept it (since I've seen different). Maybe if you explained,
: we can argue further.
Working on Wall St. there are alot of companies who spending a 1 million
or so on benchmarking is part of business. I've yet to hear any Wall St.
company have positive results with NT scaling well. If they did,
MS would be putting out ads, just like all those ads where they feature
Bankers Trust. As for Bankers Trust, here's a company who 3 years ago
refused to use C++ because MS didn't make a C++ compiler. They
went with Visual Basic. These are the companies MS likes to show off.
Yet, when companies what multiple CPU's, they have no worries using
Solaris and Sparc. Just check out the movie Toy Story, done on a
farm of Mulitple Sparc workstations running Solaris.
If JP Morgan, Merrill Lynch, Morgan Stanley, etc. had good results
with NT scaling, you'd see MS with ads talking about this. They
have not been happy with the results and that's that.
Check out this for some fun:
http://www.pcweek.com/archive/960401/pcwk0067.htm
Steve Cole
Just a note, Ed. Most of what you say has fair relevance, even if I can't
agree with all your thoughts on benchmarking nor am I a fan of NT.
However, Sun has shown that Solaris, SPARC versions at least, can be
installed on 30-way machines without modification.
Solaris, SPARC version at least, is Prego ("it's in there").
BTW, you mention that you have seen or used a Solaris box that didn't
scale well. This is interesting in the light that Byte Magazine pulled
off a multi-CPU ByteMark that scaled with 94% efficiency on a SPARC 20. I
only wish I could dig up the article (and I won't only because I broke my
ankle yesterday morn) to point at. Most tests I've seen with x86 show
40%-75% payback per processor, dependant on the OS and benchmark.
Steve Cole
On Thu, 15 Aug 1996, David Eckhardt wrote:
> There are many things that UNIX can do better, but there are many things
> that NT can do better. In a smaller orgaization (5 person) there is no
> need for UNIX unless you need a DNS for internet travel.
Why not? OK, let's put it this way:
Solaris x86: $900.00 (varies wildly for some reason)
Samba software: $0.0
GNU Compiler: $0.0
GNU Utilities: $0.0
Qualcomm POP3 Server: $0.0
INN v1.4u4: $0.0
SMTP gateway or server: $0.0
Apache Web Server: $0.0
Remote management: $0.0
POSIX compliance: $0.0
CDE: $0.0
Metadevice disk mgmt: $0.0
Remote backup: $0.0
IPX gateway: $0.0
Labour required: Depending on experience. 1/2 day fits me.
With this setup, not only do you have W3.x and W95 (and NT for that
matter) compatible file and printing facilities, you have NFS v3.0 and
full Internet compatibility right down to the mail server. You can also
use X terminals and the like for distributed calendar functionality
(something that most people overlook in Openwindows for some reason).
Price out all that functionality AND multiuser capability AND Internet
compatibility in an NT solution. And then think of scaling the solution
to 50 users. You won't likely have to change *anything* on a Solaris box.
(Linux & BSD users: yes I know the same is available, but...)
So which is more cost effective, compatible, scales better and affords
more capacity? Not NT, that's for sure!
> Larger corporations should and are using a mix of NT and UNIX. NT is a
> very good operating system for housing Applications that run on the
> Microsoft systems.
Talk to Dupont. They are replacing many of their Novell servers with NT,
but most of their attempts to migrate NT into their mainframe and UNIX
operations have failed dismally. Recently an admin at Dupont confided in
me that he has thoughts about Sun's Enterprise systems: multiple power
supplies, fail safe processors, RAID arrays etc. mean he wouldn't have
such a pressing need to carry a pager and stay close to work on weekends.
A year ago, he couldn't say enough about NT. In other words, NT failed
the test at Dupont.
> If your users are using PC's with Win95 and/or Win 3.11 then it would be
> easier to setup a Microsoft Network using NT instead of UNIX. If you
> use UNIX you will have a protocal barrier that will need NFS to cross.
Ease is relative. If you're talking about another couple of hours to set
up the server, then that's life. Otherwise, Samba provides an inexpensive
(read: free) way to provide NFS-like capabilities to Windows machines.
NFS is, however, something I use in-house between servers and PC-NFS
enabled systems.
Also, where did job security fly out the window suddenly? Any young buck
can figure out NT... ask some folk in the Netware industry how safe their
job feels this year...
> I am not saying to quit using UNIX because UNIX does have its place, but
> NT is a very good and easy to use system if you are dealing with
> computers with a Microsoft OS.
Agreed. I just like to over-design something rather than under-design it,
especially if cost factors are improved by over-design. I use strictly
Xylogics, Sun, 3Com, Motorola and Compaq - not always the performance
leaders but almost always proven, consistent and well-made... I like my
software the same way. Big companies still use UNIX for their core
information and have for 15 and more years - who am I to disagree?
> UNIX and NT are both very good and strong operating systems.
With reservations, I agree. I've picked sides and am developing as much
information and credentials as I can muster to specialize in UNIX.
Bryan Althaus
: >: br...@panix.com (Bryan Althaus) wrote:
: >:
: >: >It's been said over and over and even MS admit that after 4 CPU's,
: >: >NT does not scale well and they are working on it. It just happens
: >: >that 4 CPU's is the limit for Intel CPU's which is what 99% of NT
: >: >system run on. Digital is a minor player and in two years when
: >: >its chapter 11 and bought by Sun no one will remember the company.
: >:
: >: Show me WHERE EXACTLY SOMEONE TESTED more than 4 CPU's and SHOWED THAT NT
: >: FELL OFF DRASTICALLY!!!!! Until you come up with more than that it's been
: >: said crap, I'm just not buying it. I've seen results of 8 CPU systems, so
: >: perhaps I'm not quite as easy to convince.
: >:
: >First don't scream. Second unless you live under a rock, every major
: >computer publication has said NT does not scale well beyond 4 CPU's.
: >I guess there wrong and your right. This was for NT 3.51 and things
: >will get better.
:
:
: Show me (yes I'll scream until your thick skull gets this through it) 1
: PUBLICATION that TESTS NT on greater than 4 CPU systems!!! Show me NOW!
Your not going to start crying are you? I read many magazines and analysts
are always quoted as saying NT does not scale well. These opinions are
not based on PCWeek or BYTE or some of those other magazines you read
doing some lame benchmark, there based on real world companies doing their
own in house benchmarks. The result is NT does not scale well over 4 CPU's.
You can believe what you and MS want to believe, the rest of us will go
with what others have found out the hard way.
Ling Wang
Edmond wrote:
>
> ste...@apertus.com (Steve Kappel) wrote:
> >Like it or not, SPECrate for a 24 processor Solaris/SPARC machine is 41967 and
> >a 48 processor is 82522 - a ratio of 1.97 (with 2.0 being perfect).
> >
>
> I don't think your understanding the result. SPEC is not internally meant to be a
> multi-CPU benchmark. Vendors are using special compilers that are looking for
> parallel sequences and actually changing the behavior of the benchmark. You again
> ignored the issues I addressed with that. You also have NO PARALLEL to NT running
> SPEC with the same compiler. Even if you did, "The results for SPEC wouldn't
> change by running it under Windows NT".
SPECrate is not intended to utilize multithreaded compilers. It is the
rate the jobs finishes of, obviously if you feed it a similar number of
jobs as there is CPU's, the there would be little requirement for parallism
at all.
Go look at the threaded SPECint/rate results. That is purely an compiler benchmark.
>"SPEC was written in such a way that it's
> platform/OS independent", Jeff Reily, from SPEC org from comp.benchmarks. He then
> went on to say that you'd only be testing the differences between compilers by
> switching from UNIX to NT. SPEC is thus a hardware benchmark. You would have
> gotten the same ratio with any other OS on that same hardware platform. I'm not
> even sure about the result as it is since the benchmark changes when compiled in
> that way.
Tell that to an OS scheduling 64 concurrent active processes and distributing
it to 64 processors all sharing the same bus and memory, but different cache.
Go run SPECrate on an SunOS box and tell me the result.
> Recall that a year ago, we found out that Intel's SPEC92 results were off. It's
> not SPEC's fault, but vendors trying to squeeze the best result out any which way
> they can. I don't even think you can compare that result to anything. I'm still
> not even sure how it would show that Solaris scales. What goes into OS scaling
> (yes here we go again with this one)? Which part of scaling belongs to the OS
> (kernel level) and which part to the application (user level)? Which part
> determines how well an OS scales? Does the algorithm used have anything to do
> with how well something scales? (I know the answer to these, but I think you need
> to answer them for yourself).
> >You complain about not seeing results that *prove* NT doesn't scale well. Yet
> >you are unable to produce results that prove NT *does* scale well. If so many
>
> You need to read the questions I wrote above before we can go on here. Testing
> how well the OS ALONE scales is difficult quite honestly. There are so many other
> factors involved when benchmarking.
But who creates systems that the OS cannot handle. I don't see cray making
a CS256000 w/16XDbuses because they would be selling snake oil if they
just dropped the standard OS into it. Sun had the SC2000 go up to 16 cpus
because that is what they thought was the practical limit, then later
upped it to 20. I don't know if other vendors are like that.(besides those
who sell you pentiums to run the DOS microkernel)
> >people are doing this and the results are so good, why the hell don't they
> >publish it all over the place? My common sense says there is a reason why
> >they don't publish it. I've given you a solid benchmark result that would cause
> >any rational person to conclude that Solaris/SPARC scales _very_ well. Show
> >me any kind of broad NT benchmark that supports your opinion. How about
> >Oracle TPC results?
> >
>
> People aren't doing anything except talking at the mouth such as yourself. No one
> is proving anything because they don't have the data to. You STILL haven't come
> up with the quotes where people are saying these things all over the place. I've
> read what they've said in PC Week and InfoWorld and gave my reasons why.
The Wintel guys out there wants to benchmark on SI, sysmark, PCstone, whatever.
They want to live in the world of how fast their PCs are compared to the
REFERENCE: IBM PC 8088.
> >>Show me (yes I'll scream until your thick skull gets this through it) 1
> >>PUBLICATION that TESTS NT on greater than 4 CPU systems!!! Show me NOW!
> >>Show me the evidence of such. Show me where they said it and how they tested
> >>this. I'm not living under a rock, I've seen the publications. 1 quote from
> >>PC Week quoted a manufacturer of the new 8 CPU systems that said scalability
> >>would drop off after 8 (gosh do you think they said this to sell their
> >>technology). I'm sure you could find a manufacturer who sold a 4 CPU system
> >>that said that it falls off after 4. You could find a 16 CPU manufacturer
> >>that said it falls off after 16 and you could find a 32 CPU manufacturer that
> >>would say it falls off after 32. Which would you find more of? Which had a
> >>definitive test to show this? I'm not asking too much at all when I say you
> >
> >I wouldn't rely on a "Publication" for benchmarking large-scale
> >systems. They simply wouldn't have a clue how to do it. "PC Week"? Now
> >there's an experienced large-systems bunch!
> >
>
> PC Week simply said that it was unproven. They did no benchmarking. Few have and
> you have no results from any. They quoted a manufacturer of Corollary's new C-Bus
> II architecture (an 8 CPU architecture) that said that NT wouldn't scale past 8
> processors. The manufacturer doesn't even make a 16 processor system nor is there
> a specification for one in the x86 world. Some vendors like Sequent, AST (I
> believe or it's another acronym that starts with 'A'), have built their own
> proprietary architectures and thus had to do the porting themselves. Again, MS
> can't address every architecture in NT. You've got to set a target. What scales
> well for one may not scale well for another. The proprietary vendors had to
> ensure that it scaled on their architectures by making changes themselves.
PC mags are the bible for all those pseudo knowledgeable people in the IS
world. I know of so many IS people who tells me beefs about UNIX/RISC not
much after those mags wrote about it. The screwed up thing is
that so much people make their decision on those crap. You see those
guys salavating every time a issue comes w/the headline: "200Mhz
pure muscle.", while everyone was yawing when they published
300Mhz Alphas(Those mags published because it run NT). Then
they test the boxes with some off the shelf stuff optimized for
Intel, and the publisher yawns too.
> You really don't understand benchmarking at all. That's my biggest beef with you.
> You come off with too much talk and don't understand the underlying issues.
> Answer the questions I've drawn out for you before. You may learn much in the
> process.
> >OK, show me the data for 1 32 processor NT system! Come on now, it supports
> >32, right? Your vapor is no less so than OS/2. Go get the damn SPECrate
> >table and you will see numbers on Solaris/SPARC ALL THE WAY TO 64 PROCESSORS.
> >Come on! Digital has big monsters. Sequents run NT? NCR? Come on, where
> >is the graph from ANY of these vendors? If I were a marketing guy in any
> >of these companies I'd be splattering that data in Time magazine if it was
> >good!
> >
>
> Where is the parallel compiler for x86 NT based systems?
What's with this parallel compiler again? Parellel compilers are for
The SP2, ncube, T3Es, SPECrate is not SPECint/fp which
takes slight advantage of parallism. We were talking about the OS
being able to juggle 64 bowling balls at the same time and still
able to blink the cursor at you.
Just go write a program that does some bogus calcuation on a big set of numbers,
and just keep running more copies as you plug in more CPUs, then you could
see through that scalability haze.
> Where is the 4 CPU Intel systems of any kind? Where is the 2 CPU systems of Intel
> systems of any kind?
Hardware mfg utilizes the spec suite when designing their product. You think
Intel gives their mboard to Joe Schmo mail order and tell them to benchmark
the board with spec? spec numbers are out even before the systems or cpus are out.
> How does SPEC show how well the OS scales when even SPEC
> admits that it doesn't. You can jump onto comp.benchmarks any time and tell
> everyone that SPEC tests how well the OS scales.
>
you post this in comp.benchmark and people will respond: "But does the Cray
6400 smoke on Quake?" The only scaling that goes on in that newsgroup
is MHZ/fps scaling.
On the other hand, maybe this is where this thread belongs to. Leave the comp*unix*
alone, we don't have to hear it from you guys about how good NT is compared to UNIX.
we could just read those PeeCee mags. You probably could leave SCO on the newsgroup,
since they sold out to the devel already. ;->
Edmond
Ling Wang <wang...@jpmorgan.com> wrote:
>SPECrate is not intended to utilize multithreaded compilers. It is the
>rate the jobs finishes of, obviously if you feed it a similar number of
>jobs as there is CPU's, the there would be little requirement for parallism
>at all.
>Go look at the threaded SPECint/rate results. That is purely an compiler benchmark.
>
I will argue with you for now on as you seem to want to post intellectual
material. I agree with all of the above. If you look at how SUN and DEC
developed SPEC for their SMP machines, you'll see what I'm talking about.
>
>>"SPEC was written in such a way that it's
>> platform/OS independent", Jeff Reily, from SPEC org from comp.benchmarks. He then
>> went on to say that you'd only be testing the differences between compilers by
>> switching from UNIX to NT. SPEC is thus a hardware benchmark. You would have
>> gotten the same ratio with any other OS on that same hardware platform. I'm not
>> even sure about the result as it is since the benchmark changes when compiled in
>> that way.
>
>Tell that to an OS scheduling 64 concurrent active processes and distributing
>it to 64 processors all sharing the same bus and memory, but different cache.
I'm not sure what your getting at here. Come at me again.
>> >You complain about not seeing results that *prove* NT doesn't scale well. Yet
>> >you are unable to produce results that prove NT *does* scale well. If so many
>>
>> You need to read the questions I wrote above before we can go on here. Testing
>> how well the OS ALONE scales is difficult quite honestly. There are so many other
>> factors involved when benchmarking.
>
>But who creates systems that the OS cannot handle. I don't see cray making
What do you mean by THE OS? There are many OS's? People create systems with
OS's in mind.
>a CS256000 w/16XDbuses because they would be selling snake oil if they
>just dropped the standard OS into it. Sun had the SC2000 go up to 16 cpus
>because that is what they thought was the practical limit, then later
>upped it to 20. I don't know if other vendors are like that.(besides those
>who sell you pentiums to run the DOS microkernel)
>
Your absolutely right. People develop to what hardware they feel their systems
will run on. This was my complaint about OS/2 2.11 SMP. There wasn't any
hardware to test it's scalability claims. Recall that Sequent and the others
had to redesign the HAL under NT to fit their hardware. They may have also
changed small portions of the kernel to suit their needs. One architecture can
be much different than another. One runs best on 1 may not be best for the
other. SUN has the advantage of making Solaris for SUN systems. Let's see
Solaris on that Sequent box.
>PC mags are the bible for all those pseudo knowledgeable people in the IS
>world. I know of so many IS people who tells me beefs about UNIX/RISC not
>much after those mags wrote about it. The screwed up thing is
>that so much people make their decision on those crap. You see those
>guys salavating every time a issue comes w/the headline: "200Mhz
>pure muscle.", while everyone was yawing when they published
>300Mhz Alphas(Those mags published because it run NT). Then
>they test the boxes with some off the shelf stuff optimized for
>Intel, and the publisher yawns too.
>
I couldn't agree more. I've made this same argument for the Alpha platform
myself. Not only are they optimized for x86, but in some cases they even work
differntly (meaning it wouldn't be the same, parallel, benchmark).
>>
>> Where is the parallel compiler for x86 NT based systems?
>
>What's with this parallel compiler again? Parellel compilers are for
>The SP2, ncube, T3Es, SPECrate is not SPECint/fp which
>takes slight advantage of parallism. We were talking about the OS
>being able to juggle 64 bowling balls at the same time and still
>able to blink the cursor at you.
>
This isn't what people are comparing online though.
>Just go write a program that does some bogus calcuation on a big set of numbers,
>and just keep running more copies as you plug in more CPUs, then you could
>see through that scalability haze.
>
If scalability to you is the capability to task across multiple CPU's, then the
argument is over. One can certainly (I can certainly prove this one) "juggle"
32 tasks and have them execute much quicker on 32 processor machines running NT
than 16 or 8 processor machines. I don't know necessarily if this is OS
scalability. After all, a lot is at the discretion of the programmer. What
percentage of the overall rate goes to the OS and what percentage goes to the
processor? Does one count the amount of time priveleged code takes to execute
vs. non priveleged code vs. idle processor time and distribute scalability
between the 3 factors? It's really a tough call quite honestly. We are,
recall, talking about OS scalability.
>> Where is the 4 CPU Intel systems of any kind? Where is the 2 CPU systems of Intel
>> systems of any kind?
>
>Hardware mfg utilizes the spec suite when designing their product. You think
>Intel gives their mboard to Joe Schmo mail order and tell them to benchmark
>the board with spec? spec numbers are out even before the systems or cpus are out.
>
I was just saying that we don't have parallel comparisons here. I am correct
on that point.
>On the other hand, maybe this is where this thread belongs to. Leave the comp*unix*
>alone, we don't have to hear it from you guys about how good NT is compared to UNIX.
>we could just read those PeeCee mags. You probably could leave SCO on the newsgroup,
>since they sold out to the devel already. ;->
The last isn't my intent at all. Quite honestly, I use UNIX a lot. I use NT a
lot also. I was just setting the record straight about scalability, reality,
etc.
Edmond
br...@panix.com (Bryan Althaus) wrote:
>Your not going to start crying are you? I read many magazines and ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Has the argument come to that? I guess you have nothing else.
>analysts
>are always quoted as saying NT does not scale well. These opinions are
>not based on PCWeek or BYTE or some of those other magazines you read
>doing some lame benchmark, there based on real world companies doing >their
>own in house benchmarks. The result is NT does not scale well over 4 CPU's.
>
>You can believe what you and MS want to believe, the rest of us will go
>with what others have found out the hard way.
Which systems with more than 4 CPU's were they using? How did they test
OS scalability? I'm sorry, but your not getting anywhere with this in
house thing. If the rest of everybody goes along with that, their
gullable creatures indeed. Just to point out to some of those getting a
little too angry: I'm not attacking UNIX in any way, shape, or form.
I've said nothing bad about it other than the truth. Solaris x86 didn't
scale linearly and neither did it on 2 SS20's running version 2.5.1 doing
anything. This is not to say that the SS20's didn't scale well (they
wouldn't be used if it didn't), just putting things a little more
realistically that's all.
Also, there are in house companies everywhere which will have differing
results than you. I guess it would depend on what they were doing and
what systems they were running. Without knowing what they were doing, I
just can't accept it (since I've seen different). Maybe if you explained,
we can argue further.
-----------------------------------------------------------------------
John Wiltshire
Jouko Holopainen <jouko.ho...@xnet.otm.fi> wrote:
>John Wiltshire wrote:
>> >It might be a modern product in the kernel but it is still crippled by
>> >a poor API and user environment.
>>
>> Why is it a poor API? Seriously, as an NT developer I'd like to know
>> what I am missing out on.
>POSIX. You either get native win API (with GUI etc.) OR very limited
>subset of posixes.
You don't have to use the GUI with Win32. You can either compile a
'windows' application or a 'console' application - both are supported
by the system.
I know that NT only supports POSIX.1 and that was pretty much so the
US government could buy it.
>You cannot use both, because posix is separate subsystem, it is
>not atop of NT API.
Personally (and I know this is subjective), I find that the Win32 API
is at least equal to the Posix API in the systems I have developed.
What does Posix have that Win32 doesn't?
>> NT is slowly growing up to the higher limits of performance. I've
>> heard that there is currently no point running more than 8 CPUs but it
>> scales OK up to that (32 CPUs are supported by the kernel at the
>> moment).
>So why it is not a multiuser OS?
>This just means that I cannot compile (or access anything
>big) over modem.
>Please, don't tell me about 3rd party products, this
>really should be in the kernel itself.
Well, not the kernel, but shipped with the OS I'm sure you mean! I'd
rather have as many daemons (services on NT) running as a well-defined
user, not a superuser.
This is a falling down on NT when looked at from the Unix world, but I
personally find it not so much of a hassle (probably just I don't know
what I'm missing?)
John Wiltshire
-----------------------------------------------------------
John Wiltshire | Email j...@qits.net.au
Network Systems Engineer | Ph. W +61 7 38342783
Traffic Systems Development | H +61 7 33006052
-----------------------------------------------------------
These opinions may not be mine and are almost certainly
not those of my employer!
Saint (n) -
A dead sinner, revised and edited.
Jianlin Chang
It is hard to say, DEC UNIX is not performing well, it's not certain
how well they will perform with Alpha NT. I can't see DEC's strong
commitment to UNIX.
Wayne Hyde (w...@cis.ufl.edu) wrote:
: Bryan Althaus <br...@panix.com> wrote in article <4ut4ha$5...@panix.com>...
: > Shawn D. McPeek (smc...@coewl.cen.uiuc.edu) wrote:
: > : I don't believe you need to worry about Digital going chapter 11, it's
: > : doing quite well.
: > Hello, they just lost ~$300 million last quarter. It's the first lose
: > in about 2 years. There PC division is doing badly, and they have
: > no quick fix yet.
: And IBM lost over 6 _BILLION_ a few years back. DEC is a big company.
: They will survive a 300M loss.
: --
:
Evan Leibovitch
>>> If your users are using PC's with Win95 and/or Win 3.11 then it would be
>>> easier to setup a Microsoft Network using NT instead of UNIX. If you
>>> use UNIX you will have a protocal barrier that will need NFS to cross.
>>Not any longer! See SCO's VisionFS announcement last week. It requires
>What about Samba?
Samba is for those who can't afford VisionFS :-).
--
Evan Leibovitch, Sound Software Ltd., located in beautiful Brampton, Ontario
Caldera Business Partner / SCO Authorized VAR / ev...@telly.org / (905) 452-0504
I love standards because they give non-conformists something not to conform to
Ling Wang
Bryan Althaus wrote:
>Working on Wall St. there are alot of companies who spending a 1 million
>or so on benchmarking is part of business. I've yet to hear any Wall St.
>company have positive results with NT scaling well. If they did,
>MS would be putting out ads, just like all those ads where they feature
>Bankers Trust. As for Bankers Trust, here's a company who 3 years ago
>refused to use C++ because MS didn't make a C++ compiler. They
>went with Visual Basic. These are the companies MS likes to show off.
MS really exploited it when Merrill Lynch made a mega million deal
with them. There was some major gonad kissing going on there(Hey,
even when buying cheap crap like $50k raid boxes, the brown nosing
sales reps are all over you. And when you bring up issues like
competetor's features and price, its like nobody even cares about
that because the bribe has taken place already.)
After that, it's like follow the leader by other wall street weenies
& wannabees.
>Yet, when companies what multiple CPU's, they have no worries using
>Solaris and Sparc. Just check out the movie Toy Story, done on a
>farm of Mulitple Sparc workstations running Solaris.
I've seen computer mags talk about the computer graphics on Toy Story,
but not mention 1 word on the system & software. Probably because
if it ain't a PeeCee, it ain't of interest to us(because we can't tell
you how many quake fps it does)
>If JP Morgan, Merrill Lynch, Morgan Stanley, etc. had good results
>with NT scaling, you'd see MS with ads talking about this. They
>have not been happy with the results and that's that.
It is not whether NT scales well or not for the corporate database or
the giant compute server farms, it seems to be more of the market
demand, and maybe some dopey IS execs getting their facts from
infoweek(formerly Open System Today?, & before that UNIX Today?)
And their even dopier Devry Institute graduates subordinates reading
those PeeCee rags and telling them what to do. Those screwdriver
carrying guys thinks when NT scales to 4CPU, its time to unplug the
Tandems and Amdahls. Hey with 8CPUs, unplug the Crays, iPSCs, CM5s.
Global enterprise here we come(they don't know anything about these
machines, they just pass by these giant boxes in the datacenter when
they are going to their PeeCee server to copy something onto the
floppy or do a CTRL-ALT-DEL.)
BTW, whats my whole response got to do with scaling? :-o
Danny Aldham
X-Newsreader: TIN [version 1.2 PL2]
John Wiltshire (j...@qits.net.au) wrote:
: Hmm..
: NT Server 4.0 $500 (or so, I'm from Australia so I can't quote US)
: TCP/IP, NetBeui,IPX $0
: GNU Compiler $0
: Mail Server $0 (plenty of them out there)
: IIS Web Server $0
: Remote Management $0
: None of the other stuff costs anything either (from my experience), so
: I won't go on.
And I understand they are going to include a command line or shell in
NT4.0 Include a hobbled perl5 and an add-on telnet $$$ and you almost
have a Unix box. But I have to call bullshit on the $500 dollar cost.
Danny Aldham
Danny Aldham
X-Newsreader: TIN [version 1.2 PL2]
Jianlin Chang (chang@mjs1) wrote:
: It is hard to say, DEC UNIX is not performing well, it's not certain
: how well they will perform with Alpha NT. I can't see DEC's strong
: commitment to UNIX.
I don't see any commitment by DEC to unix, but everyone is talking about
how NT doesn't scale beyond 4 or 5 Pentuims. I think it would be a
screamer on 4 or 5 Alpha's.
Danny Aldham
Steve Kappel
In article <4uvufe$6...@lace.colorado.edu>, Edmond <unde...@rmii.com> writes:
>ste...@apertus.com (Steve Kappel) wrote:
>>In article <4uve9s$r...@lace.colorado.edu>, Edmond <unde...@rmii.com> writes:
>>>ste...@apertus.com (Steve Kappel) wrote:
>>>
>>>>SPECrate is and you will find that it is a good general-purpose multiuser/
>>>>multiprocessor benchmark of real-world usage. Clearly you are more interested
>>>>in getting the answer you want - if not then the benchmark is no good.
>>>
>>>Your losing it here. I don't think any aircraft made will reach the speed of
>>>light no matter what benchmark said it did. I have reason to doubt such.
>>
>>Like it or not, SPECrate for a 24 processor Solaris/SPARC machine is 41967 and
>>a 48 processor is 82522 - a ratio of 1.97 (with 2.0 being perfect).
>>
>
>I don't think your understanding the result. SPEC is not internally meant to be a
>multi-CPU benchmark. Vendors are using special compilers that are looking for
>parallel sequences and actually changing the behavior of the benchmark. You again
>ignored the issues I addressed with that. You also have NO PARALLEL to NT running
What is your problem with automatic multithreading compilers? Cray has been
doing that for a LONG time. We are talking about the scalability of an O/S which,
to me, means how efficiently it reduces the overall amount of time it takes to
complete a set of work as you add processors. Use whatever tricks work as long
as the benchmark is doing useful real-world application loads.
You are not reading what I said. I said "SPECrate" and you keep saying "SPEC".
Get a clue, buddy! Look at http://www.specbench.org/osg/cpu92/specrate.txt .
I quote from it:
"The SPECrate is a capacity measure. It is not a measure of how fast
a system can perform any task; rather it is a measure of how many of
those tasks that system complete within an arbitrary time interval.
To use an analogy, imagine that this is not about computers, it is
about cooking stoves. The SPECmark89 would be a rating of how fast
one burner could bring one cup of water to a boil and the SPECrate
would be rating of how much water can be boiled by that stove (using
whatever number of burners available). If a simple single-burner
stove could boil one cup of water in 5 minutes, then it would have
a rate of 12 cups an hour; but if a four-burner stove could bring
four separate cups of water to boil all in 15 minutes, it would
have a rate of 16 cups an hour. The second stove would have a
greater capacity, i.e. a higher SPECrate, but the first stove
would be better for boiling an individual egg due to its better
SPECmark89."
Perhaps your fundamental blockage about this is that the entire concept
of an SMP system doing a lot of diverse things for many users ALL AT ONCE
is completely foreign to Micro$oft advocates.
Am I to blame for there being "NO PARALLEL to NT"? No. Oracle runs on NT.
Oracle benefits from SMP architectures. Don't tell me that somebody couldn't
run a TPC database benchmark against an NT SMP machine and SHOW US THE
CURVE! Isn't an NT server providing database functionality for a network of
PCs an example of real-world application loads?
>SPEC with the same compiler. Even if you did, "The results for SPEC wouldn't
>change by running it under Windows NT". "SPEC was written in such a way that it's
>platform/OS independent", Jeff Reily, from SPEC org from comp.benchmarks. He then
>went on to say that you'd only be testing the differences between compilers by
>switching from UNIX to NT. SPEC is thus a hardware benchmark. You would have
Hmmm... Your logic is full of craters. "SPEC is thus a hardware benchmark." So
running Unix and NT on the same box and measuring the time it takes to do some
real-world application will be exactly the same? No, it won't be. The sentence
just before that quote says "you'd only be testing the differences between
compilers". So now SPEC on the same hardware is showing us how good the
compilers and _entire-environment-including-os_ are at solving the problem.
I, and most people, care about HOW LONG IT TAKES TO GET THE JOB DONE. If
one compiler/os verses another gets it done in less time then it sure looks
*better* to me if performance is my #1 concern (granted that it usually isn't
the only concern).
>>You can bitch about what SPECrate is all you want but it is a "good" benchmark.
>>None are perfect. The fact is that in large configurations the ratio is almost
>>linear.
>
>Recall that a year ago, we found out that Intel's SPEC92 results were off. It's
>not SPEC's fault, but vendors trying to squeeze the best result out any which way
>they can. I don't even think you can compare that result to anything. I'm still
>not even sure how it would show that Solaris scales. What goes into OS scaling
>(yes here we go again with this one)? Which part of scaling belongs to the OS
>(kernel level) and which part to the application (user level)? Which part
>determines how well an OS scales? Does the algorithm used have anything to do
>with how well something scales? (I know the answer to these, but I think you need
>to answer them for yourself).
I don't care *where* scaling comes from. It is simply a matter of cost/benefit
of getting whatever work done that needs to get done. Lets say I have a huge
Oracle database and I'm going to initially buy an 8 processor machine.
Workload almost always grows with time. Lets say I need to support 2X as many
users in a year. If I can add ~9 processors to support that under one O/S and
I have to add 12 processors under another then I say that the later is not as
scalable and it costs me more money to do the same exact work. Why is that
such a hard concept to understand?
>>You complain about not seeing results that *prove* NT doesn't scale well. Yet
>>you are unable to produce results that prove NT *does* scale well. If so many
>
>You need to read the questions I wrote above before we can go on here. Testing
>how well the OS ALONE scales is difficult quite honestly. There are so many other
>factors involved when benchmarking.
The O/S is a major contributor to scalability. You are too tied up in splitting
hairs of "true benchmarking". All I care about is cost/benefit of getting real-world
workloads done.
I would be very surprised if SunSoft doesn't use things like SPECrate to test
new versions of Solaris verses old. If those numbers got worse I'd sure as hell
say that something is wrong. If they improve then somebody is doing things
right. The better they are at SMP the greater their advantage in the marketplace.
>People aren't doing anything except talking at the mouth such as yourself. No one
>is proving anything because they don't have the data to. You STILL haven't come
Solaris has the data. I'm still waiting for NT data. As far as I'm concerned, the
absence of published data for a mature product is as good as meaning
the data isn't good.
Naw, if I were an NT SMP vendor I wouldn't need to publish my great scalability
numbers. I can keep selling this purely on M$-driven hype. Marketing-types
love benchmarks when they are good and hate them when they are bad.
>up with the quotes where people are saying these things all over the place. I've
>read what they've said in PC Week and InfoWorld and gave my reasons why. These
>quotes were from specific manufacturers who produced SMP hardware running NT.
>NT's scaling broke off where their specifications did. In each case, they were
>different. Is Oracle designed to run on NT systems with more than 8 processors?
>I'd like to here from an Oracle representative about that one.
>For instance, SQL Server isn't designed to run on systems with more than 6
>processors since the other folks who were using it on their machines were at most
>using the 6 processor systems from Intergraph. I do know of results for greater
>than 8 CPU systems (Hmmm... now how would I get those). They have little to do
>with NT's ability and more to do with the hardware though. Their also at the
>discretion of the manufacturer and not my own. I wouldn't need them to end this
>argument anyway as you have no proof of either.
Given the following facts:
1. Oracle does support >8 processors on Unix machines.
2. NT supports 32 processors.
3. Many NT vendors have hardware >8 processors.
So why haven't we seen any benchmark #s. Gee, maybe it is because they suck!
>proprietary architectures and thus had to do the porting themselves. Again, MS
>can't address every architecture in NT. You've got to set a target. What scales
>well for one may not scale well for another. The proprietary vendors had to
>ensure that it scaled on their architectures by making changes themselves.
Interesting. I have a SPARCstation IPC at home that is, what, maybe 6 years
old. About a PC-AT by comparison to the machines today. The same Solaris
that runs well on that runs on multimillion-dollar Enterprise SMP servers.
To be sure there are several generations of fundamental architecture
differences between that IPC and an Enterprise server. Sun doesn't seem to
have any problem scaling well on all of them.
>You really don't understand benchmarking at all. That's my biggest beef with you.
>You come off with too much talk and don't understand the underlying issues.
>Answer the questions I've drawn out for you before. You may learn much in the
>process.
Ah... the benchmarking is a religion thing. I don't run benchmarks on my
machines day-in-day-out but when they provide data that mirrors what I
experience doing my day-to-day job then they are valuable to me in making
decisions. That is all I care about.
Edmond
>: what systems they were running. Without knowing what they were doing, I
>: just can't accept it (since I've seen different). Maybe if you explained,
>: we can argue further.
>
>or so on benchmarking is part of business. I've yet to hear any Wall St.
>company have positive results with NT scaling well. If they did,
>MS would be putting out ads, just like all those ads where they feature
>Bankers Trust. As for Bankers Trust, here's a company who 3 years ago
>refused to use C++ because MS didn't make a C++ compiler. They
>went with Visual Basic. These are the companies MS likes to show off.
>
Did those companies test systems running NT on greater than 8 CPU systems?
I'll ask Sequent myself and see if this is true. If not, then again, you
have no parallel.
>http://www.pcweek.com/archive/960401/pcwk0067.htm
>
You still have no direct comparison to Solaris or another UNIX system running
on the SAME hardware using the same benchmark and compiler technology. All
you had was Netware vs. NT using that benchmark. It also didn't test NT on
higher order systems. I've already told you I've read those results before
anyway. Here's what you need to show:
NT running on the same hardware, running the same test, using the same
compiler technology vs. the others. Without that, one cannot be sure what's
occuring. Again, I've seen things for myself and understand what NT can do.
Edmond
Steve Cole <co...@post.kosone.com> wrote:
>I know this is hard for you, Edmond. But shouldn't every compiler which
>runs on multi-processing machines actually compile the code as efficiently
>for parallel processing as possible? What it does for benchmarks it can
>do for other code, right?
>
You answer this question. What's your take on this?
>In other words, are you saying that NT's only problem is that it doesn't
>have a good compiler? Or are you saying that compilers on everything but
>NT are so clever that benchmarks mean nothing anyways?
>
First of all, you don't even have a comparison that I can lay out. I don't
know of any NT compilers that look for parallelism in your code and exploit
that feature on SMP machines. Basically, I don't know of any NT compiler
that is specifically designed for SMP machines. Do they need to be? Not if
you know what your doing.
Edmond
>http://www.pcweek.com/archive/960401/pcwk0067.htm
>
-----------------------------------------------------------------------
John Wiltshire
br...@panix.com (Bryan Althaus) wrote:
>Your not going to start crying are you? I read many magazines and analysts
>are always quoted as saying NT does not scale well. These opinions are
>not based on PCWeek or BYTE or some of those other magazines you read
>doing some lame benchmark, there based on real world companies doing their
>own in house benchmarks. The result is NT does not scale well over 4 CPU's.
>You can believe what you and MS want to believe, the rest of us will go
>with what others have found out the hard way.
As one of NT4's sales cries is that it scales well to double the
number of CPUs for DB serving (from 4 to 8), I'd be confident in not
recommending more than 8 CPUs on an NT machine. I tend to listen when
MS tells us limitations of their own systems (it doesn't happen
often).
Glenn Davies
j...@qits.net.au (John Wiltshire) wrote:
>Steve Cole <co...@post.kosone.com> wrote:
>Hmm..
>NT Server 4.0 $500 (or so, I'm from Australia so I can't quote US)
>TCP/IP, NetBeui,IPX $0
>GNU Compiler $0
>Mail Server $0 (plenty of them out there)
>IIS Web Server $0
>Remote Management $0
I find it hard to believe that Australians can buy NT server for half
the price of what Canadians pay. The Canadian and Australian $ are
roughly the same value against the US $
---
Glenn Davies inet: gl...@direct.ca
Imago Systems Inc - Virtual Reality Technology
phone: (604) 681-9288 fax: (604) 681-8705
Brandon S. Allbery KF8NH
Also sprach ev...@telly.org (Evan Leibovitch) (<Dw9CC...@telly.org>):
+-----
| Given the surprisingly-fast withdrawl of the NEC MIPS-based machines
| from the market, I wonder how *any* of the non-Intel NT engines (such
| as the Alpha) are faring.
+--->8
The long delay in MS's enabling non-Intel NT to run Office 95 in some form
could be taken as an indication... and even so, it is my understanding that
it will be emulated, not ported.
--
++brandon s. allbery flying with merlin! b...@kf8nh.apk.net
telotech's "loup-guru" :-) FORZA CREW! b...@telotech.com
Mr. Who
Danny .....
My understanding of NT is that it doesn't scale beyond 4 CPUs
regardless of which CPU it is.
Perhaps a forthcoming version of NT which will supposedly permit
clustering can resolve this shortcoming.
And, if you took a DEC Alpha running a database engine .... like
Oracle or Informix under NT Server and under Ultrix which one would
produce the higher TPC/C marks ? I'd bet on Ultrix myself.
Now, if you want proven scalability ... go with Solaris ... Sun has
demonstrated linear scalability up to 64 CPUs with their product....
Ok, you can't buy the 64 CPU machine yet.... But anyone that has a
spare $2 million or so can pick up an Ultra Enterprise Server 6000
with 30 CPUs and 30GB RAM. That should scream versus a machine with
4 or 5 Alphas.......
Just my $.02 in this discussion
Flames to /dev/null
--
+---------------------------------------+
| Mr. Who is: |
| jkot...@interaccess.com (Personally) |
| jo...@sigg.com (Professionally) |
| |
| So sayeth Dilbert: |
| Technology is no place for wimps! |
+---------------------------------------+
John Wiltshire
Steve Cole <co...@post.kosone.com> wrote:
Hmm..
NT Server 4.0 $500 (or so, I'm from Australia so I can't quote US)
TCP/IP, NetBeui,IPX $0
GNU Compiler $0
Mail Server $0 (plenty of them out there)
IIS Web Server $0
Remote Management $0
None of the other stuff costs anything either (from my experience), so
I won't go on.
You could throw in Exchange if you felt like it too.
Now, you have user profiles, integrated single logon security,
workstation independance for the users etc.
Probably about the same merits really...
Charles Hedrick
da...@hendrix.postino.com (Danny Aldham) writes:
>I don't see any commitment by DEC to unix, but everyone is talking about
>how NT doesn't scale beyond 4 or 5 Pentuims. I think it would be a
>screamer on 4 or 5 Alpha's.
huh? What sort of commitment do you want to see? DEC is continuing
to develop and support Unix aggressively. Digital Unix 4.0, which was
just released, has a number of significant improvements from previous
releases, and was specifically intended to make sure that Unix would
run on some of the lower end machine that previously hadn't been able
to run it. The high-end machines were already Unix-oriented. I'm by
no means sure that DEC's strategy of supporting 3 OS's (4 if you count
Linux) is working for them, but I see no evidence that they're
changing it. If they wanted to go to a single OS, Unix would almost
certainly have to be it, but I don't see much evidence that that's
going to happen. (They couldn't make NT their only OS, because that
would put their ability to do major new products too much under the
control of Microsoft. It would turn them into a commodity hardware
vendor. They're already seeing in the PC market how dangerous that
is. I'm not saying I think supporting NT is a mistake for them. I
don't think it is. But having it as their only OS would be a horse of
a very different color.)
Steve Cole
On Sat, 17 Aug 1996, John Wiltshire wrote:
> Now, you have user profiles, integrated single logon security,
> workstation independance for the users etc.
>
> Probably about the same merits really...
Not at all. Define your take on the terms:
---
SMTP mailer - <names of packages, costs, featureset>
GNU compatibility - are these hacked-down versions of normal GNU?
Remote administration - could you be like me and use a box 50 miles away
for *everything* you do?
CDE and/or X-Windows compatibility
POSIX Compliancy
---
Saying you have these features is like Wendy's saying to a master chef "we
have good chili". It might be good, but not equivalent to the "real
thing" in an knowledgable person's opinion!
While we're at it, let's drop a few more things on the Solaris x86 vs NT
arguement for fun:
Solaris benefits:
Applications that have had 15 years and more to mature.
Real-time capabilities with multi-threaded 64-bit kernel I/O (s2.5.1)
$130 (Can.) for the media of your choice if you have a SPARC-based system.
More than 10,000 native applications (SPARC), way more than NT!
Optimized version available for streaming video service applications.
Self-optimizing kernel (adds and deletes buffers as required, FI).
More robust memory protection than NT.
Self-optimizing UNIX file system (co-existant 8K and .5K 'block sizes')
Flat file system structure (what's with this C:, D: crap?!?!)
*Real* plugnplay on SPARC & PPC systems, NT-like P&P on x86.
Support for standards such as CORBA, NEO, JAVA without major changes.
A very active source code resource with extensive native support.
As you say, compilers that exploit parallel systems to their best ability.
Extremely good scaling support for SMP machines - proven!
Mature OS means that security flaws have been discovered and documented.
Proven reliability in high-availability and high-stress conditions.
Well, that's about it... I'm wasting a lot of time arguing my point, and I
know what OS wars usually turn out as "I know you are, but what am
I?" so I'll hopefully be able to leave it at this.
Jack Love
On 14 Aug 1996 13:43:34 GMT, ste...@apertus.com (Steve Kappel) wrote:
>In article <3210FA...@Colorado.Edu>, Edmond <unde...@Colorado.Edu> writes:
>>POSIX format (though not the 0.4 standard in and of itself). Do you
>>honestly think programming widgets is easier than Windows? Just curious.
>
>As I said earlier, I'm not a Windows programmer. I've worked with Motif/Xt for
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>a long time.
>
>For trivial stuff I suppose Windows is probably easier. For more complex things,
^^^^^^^^^^^^^^^^^^^
>like those found in commercial-grade products, Motif/Xt is far more powerful
> and any additional complexity resulting from that is minor.
On what basis do you base your 'suppositions'? You have no experience in
Windows and apparently none in the real world as you aren't able to see
that there are zillions more 'commercial' Win applications than there are
Motif.
>>> The number of processors "supported" and the number that can be
>>> effectively used are two entirely different things. Most people say 4
>>> is about the end of the road for NT, others say 8 max. The efficiency
>>> of NT drops off quickly as you add processors. This is a well-known
>>> phenomena. Solaris scales near linear - just take a look at the SPECrate
>>> numbers for a Cray (now Sun) SuperServer going from 24 to 48 processors
>>> (41967 to 82522 - a ratio of 1.97).
>>>
>>
>>I'm sorry Steve, but do you want to throw a benchmark my way? How poor is
>>the scaling of NT on 8+ processor systems? How did you test this? Nothing
>
>Another post says that the MS Web site admits that 3.51 drops off after 4
>and 4.0 after 8.
Let's try a little thought experiment.
Rush out and count the number of 48 processor systems in the world.
Does that tell you how important scaling at that level is? In terms of
total potential market share?
>>scales nearly linearly (except in some odd ideal case). There's just too
>>much overhead involved in coherency, heap allocation from different threads,
>>etc. Even if it was a shared heap, I think perfect scaling sounds too good
>>to be true. This looks much like Digital's benchmark showing how their
>>64-bit CPU makes databses 8x faster than 32-bit CPU's (the data was well
>>over 6 GB large). There is also no one UNIX. BSDi, NeXTStep, etc. don't
>
>I'd say those SPECrate numbers are pretty damn close to linear. Notice
>I didn't put out numbers for going from 1 to 2 or 2 to 3 but 24 to 48. Any
>significant degradation should easily show up at that scale.
>
>Do you understand what makes up the SPEC benchmarks? Take a look
>at http://www.specbench.org/osg . These are every-day real-world
>things. No huge address space tricks.
>
>
>--
>______________________________________________________________________
>Steve Kappel Work: ste...@apertus.com
>Apertus Technologies, Inc. http://www.apertus.com
>7275 Flying Cloud Drive Home: ska...@winternet.com
>Eden Prairie, MN 55344
>
>
Evan Leibovitch
In article <4v0o45$o...@knot.queensu.ca>,
Jianlin Chang <ch...@mjs1.phy.queensu.ca> wrote:
>It is hard to say, DEC UNIX is not performing well, it's not certain
>how well they will perform with Alpha NT. I can't see DEC's strong
>commitment to UNIX.
In an odd move that could be seen as an act of desparation, DEC has bent
over backwards to make sure that Linux was ported to Alpha. It seems that
even though a Linux port to SPARC was already underway, Digital offered
resources and help enough to make the alpha port the first non-Intel
version of Linux to be offered.
Does this constitute 'commitment'? Not IMO. It seems like a crass, but
cheap, way to increase interest in their hardware beyond what it's
getting from other circles.
Given the surprisingly-fast withdrawl of the NEC MIPS-based machines
from the market, I wonder how *any* of the non-Intel NT engines (such
as the Alpha) are faring.
--
John Wiltshire
da...@hendrix.postino.com (Danny Aldham) wrote:
>X-Newsreader: TIN [version 1.2 PL2]
>John Wiltshire (j...@qits.net.au) wrote:
>: Hmm..
>: NT Server 4.0 $500 (or so, I'm from Australia so I can't quote US)
>: TCP/IP, NetBeui,IPX $0
>: GNU Compiler $0
>: Mail Server $0 (plenty of them out there)
>: IIS Web Server $0
>: Remote Management $0
>: None of the other stuff costs anything either (from my experience), so
>: I won't go on.
>And I understand they are going to include a command line or shell in
>NT4.0 Include a hobbled perl5 and an add-on telnet $$$ and you almost
>have a Unix box. But I have to call bullshit on the $500 dollar cost.
My fault - looked at the academic price.
Retail is about $800 (5 user license which is what was specified at
the beginning).
How hobbled is the perl in NT4? Anyone got any comments?
Why do you need telnet?
John Wiltshire
Steve Cole <co...@post.kosone.com> wrote:
>On Sat, 17 Aug 1996, John Wiltshire wrote:
>> Now, you have user profiles, integrated single logon security,
>> workstation independance for the users etc.
>>
>> Probably about the same merits really...
>Not at all. Define your take on the terms:
>---
>SMTP mailer - <names of packages, costs, featureset>
Pretty limited the free ones (compared to the Solaris sendmail).
Exchange really is overkill for 5 users though. You'd be better just
running a local MS-Mail network and use the MS-Mail internet gateway
(free from the SoftLib).
>GNU compatibility - are these hacked-down versions of normal GNU?
Nope. Right from GNU themselves. VC and BC are better though.
>Remote administration - could you be like me and use a box 50 miles away
>for *everything* you do?
I though you said admin, not remote use. Weren't we talking about a
small 5 person office here?
>CDE and/or X-Windows compatibility
Why is this needed for a 5 user system?
>POSIX Compliancy
Why is this needed? We have POSIX.1 (though this is c**p I know).
Win32 does pretty much everything anyhow...
>---
>Saying you have these features is like Wendy's saying to a master chef "we
>have good chili". It might be good, but not equivalent to the "real
>thing" in an knowledgable person's opinion!
>While we're at it, let's drop a few more things on the Solaris x86 vs NT
>arguement for fun:
>Solaris benefits:
>Applications that have had 15 years and more to mature.
Can't beat this one. NT is getting better all the time but still has
a way to go in maturity. When it does though, :-)
>Real-time capabilities with multi-threaded 64-bit kernel I/O (s2.5.1)
This is needed for a 5 user office?
>$130 (Can.) for the media of your choice if you have a SPARC-based system.
media? You mean CDROM, tape etc?
>More than 10,000 native applications (SPARC), way more than NT!
Comes with the maturity of the system. For a small office, NT has all
the applications you need though (I can't imagine what you'd be left
wanting in).
>Optimized version available for streaming video service applications.
C'mon... 5 USER OFFICE! Streaming Video? MS is working on streaming
video at the moment. I've seen demos and they have a campus wired up
on it at the moment somewhere in the US. Doesn't even need an
optimised version of NT.
>Self-optimizing kernel (adds and deletes buffers as required, FI).
?? What buffers? Why?
>More robust memory protection than NT.
This I am interested in. Can you split the thread and explain for me?
>Self-optimizing UNIX file system (co-existant 8K and .5K 'block sizes')
Hmm. NTFS can't do this at the moment. Is it a transaction based
system for rollback on powerdown?
>Flat file system structure (what's with this C:, D: crap?!?!)
Well, you can just use the full flat names if you want to. This way
is shorter (and helps windows applications to run). How does WABI do
this?
>*Real* plugnplay on SPARC & PPC systems, NT-like P&P on x86.
Cool.
>Support for standards such as CORBA, NEO, JAVA without major changes.
?? NT can't support these without _any_ changes? Look again - just
add the DLLs.
>A very active source code resource with extensive native support.
I'm not sure what you mean here.
>As you say, compilers that exploit parallel systems to their best ability.
gcc? This is something that is really up to the libraries and the
programmer isn't it (unless you are looking at a higher level language
than C)?
>Extremely good scaling support for SMP machines - proven!
Again, depends on the application. NT hasn't been tested well enough
on more than about 8 CPUs for good results. The x86 architecture
kinda breaks down the performance here.
>Mature OS means that security flaws have been discovered and documented.
NT was designed from the ground up for security. _Very few_ flaws
have been discovered in the system and all those that have are well
documented (or there would be more specifics in the usenet wars). I
am on just about every NT security mailing list and I still see more
flaws in Solaris than NT coming through for all your 'mature OS'
claims.
>Proven reliability in high-availability and high-stress conditions.
Since when was the 5 user system high-stress? There are times when I
will not advocate NT, but for a small 5 user office I think it is a
good idea.
>Well, that's about it... I'm wasting a lot of time arguing my point, and I
>know what OS wars usually turn out as "I know you are, but what am
>I?" so I'll hopefully be able to leave it at this.
Don't get me wrong. NT is not a universal solution but I think there
are places where it does better than a Solaris solution. The small 5
user office is one of these in my opinion (though it depends on the
previous computing experience of the staff).
I was only talking about the 5 user situation and realise that at the
high end of performance (massively parallel and even modestly
parallel), NT falls over compared to the more mature Unixes. With
each release it gets better though - companies like Sequent have been
working hard on this.
#define PREDICTION_MODE
NT will get big in the future. It will not replace Unix but work with
Unix.
#undef PREDICTION_MODE
John Wiltshire
da...@hendrix.postino.com (Danny Aldham) wrote:
>X-Newsreader: TIN [version 1.2 PL2]
>Jianlin Chang (chang@mjs1) wrote:
>: It is hard to say, DEC UNIX is not performing well, it's not certain
>: how well they will perform with Alpha NT. I can't see DEC's strong
>: commitment to UNIX.
>I don't see any commitment by DEC to unix, but everyone is talking about
>how NT doesn't scale beyond 4 or 5 Pentuims. I think it would be a
>screamer on 4 or 5 Alpha's.
NT4 scales well to about 8 CPUs now.
John Wiltshire
b...@kf8nh.apk.net (Brandon S. Allbery KF8NH) wrote:
>Also sprach ev...@telly.org (Evan Leibovitch) (<Dw9CC...@telly.org>):
>+-----
>| Given the surprisingly-fast withdrawl of the NEC MIPS-based machines
>| from the market, I wonder how *any* of the non-Intel NT engines (such
>| as the Alpha) are faring.
>+--->8
>The long delay in MS's enabling non-Intel NT to run Office 95 in some form
>could be taken as an indication... and even so, it is my understanding that
>it will be emulated, not ported.
I've heard from some guys in Digital that the Alpha can run x86
applications (32 bit ones even) faster than a PPro of the same $$
value. This is in _emulation_!
Martin Imrisek
In message <4up9l9$h...@hobyah.cc.uq.oz.au> - j...@qits.net.au (John
Wiltshire)Tue, 13 Aug 1996 07:07:43 GMT writes:
:>
:>ste...@apertus.com (Steve Kappel) wrote:
:>
:>>In article <4une9e$j...@panix.com>, br...@panix.com (Bryan Althaus) writes:
:>>>
:>>>AS a PRO UNIX person I will say that NT being chosen over UNIX is real,
:>>>and unavoidable simple because NT is the successor to Window95. Most
:>>>corporate site are skipping on Window95 and waiting for NT 4.0 which
:>>>will come with the Window95 look. True this is the client (NT workStation)
:>>>we are talking about and not the server, but once you have NT clients,
:>>>the logical choice for most will be an NT server.
:>>>
:>>>Linux will not be hurt by this but SCO surely will. SCO's main business
:>>>is in the server area, and NT will hurt SCO's sales, though not nearly
:>>>as bad as what it will do to Netware.
:>>>
:>>>For those who have never used NT, it's a nice product mainly because it
:>>>was developed by mostly x-Digital top rank and file developers who
:>>>Gates pulled in. No way MS could have developed NT on their own. Just
:>>>looked at Windows95 for an example of a MS developed OS.
:>
:>>It might be a modern product in the kernel but it is still crippled by
:>>a poor API and user environment.
:>
:>Why is it a poor API? Seriously, as an NT developer I'd like to know
:>what I am missing out on.
:>>>Though, you will notice NT more and more is trying to become like
:>>>Windows 95 by brining everything into the kernel for speed, allowing
:>>>access to devices that it once didn't, making it faster and more
:>>>useable but at the expensive of stability.
:>
:>>I've read that microkernel architectures, while elegant, don't perform or
:>>scale well. The only major Unix vendor with such a kernel is Digital.
:>>They know they have to do something with NT to fix the scaling problems.
:>
:>NT is slowly growing up to the higher limits of performance. I've
:>heard that there is currently no point running more than 8 CPUs but it
:>scales OK up to that (32 CPUs are supported by the kernel at the
:>moment).
:>DEC is working on a 64bit addressing system for NT AFAIK to help with
:>the huge DB support (and to take advantage of their Alphas).
DEC is also advertising that Linux (64 bit) exists for their Alphas!!!
:>
:>John Wiltshire
:>
:>-----------------------------------------------------------
:>John Wiltshire | Email j...@qits.net.au
:>Network Systems Engineer | Ph. W +61 7 38342783
:>Traffic Systems Development | H +61 7 33006052
:>-----------------------------------------------------------
:>These opinions may not be mine and are almost certainly
:>not those of my employer!
:>
:>Saint (n) -
:> A dead sinner, revised and edited.
:>
Martin Imrisek
Math & Comp Sci
University of Toronto
imr...@interlog.com
a378...@cdf.toronto.edu
Sangria
>>Hmm..
>
>>NT Server 4.0 $500 (or so, I'm from Australia so I can't quote US)
>>TCP/IP, NetBeui,IPX $0
>>GNU Compiler $0
>>Mail Server $0 (plenty of them out there)
>>IIS Web Server $0
>>Remote Management $0
>
>I find it hard to believe that Australians can buy NT server for half
>the price of what Canadians pay. The Canadian and Australian $ are
>roughly the same value against the US $
Check your client access licences.
Your $1000 copy of NT should include 10 client access licenses.
THe $500 (which is pretty low--I got mine for $600) doesn't
include any client licenses.
-- Sang.
********************************************************
* Sang K. Choe san...@inlink.com *
* http://sangria.inlink.com/index.html *
* finger: sa...@sangria.inlink.com *
********************************************************
John Wiltshire
gl...@direct.ca (Glenn Davies) wrote:
>j...@qits.net.au (John Wiltshire) wrote:
>>Steve Cole <co...@post.kosone.com> wrote:
>>Hmm..
>>NT Server 4.0 $500 (or so, I'm from Australia so I can't quote US)
>>TCP/IP, NetBeui,IPX $0
>>GNU Compiler $0
>>Mail Server $0 (plenty of them out there)
>>IIS Web Server $0
>>Remote Management $0
>I find it hard to believe that Australians can buy NT server for half
>the price of what Canadians pay. The Canadian and Australian $ are
>roughly the same value against the US $
My fault - quoted Academic price. US$809 for 5 user license.
Danny Aldham
John Wiltshire (j...@qits.net.au) wrote:
: How hobbled is the perl in NT4? Anyone got any comments?
Perl is hobbled from the point of view that you do not have access to
any unix systems calls. So perl code that would work on Solaris, AIX,
SCO, HP-UX, SGI, etc, etc will not run on NT. Something as simple as:
open(MAIL,"|/usr/bin/mail") || die ; will die . The most common question
at comp.lang.perl.misc is How do I do XXXX in NT ?
: Why do you need telnet?
I suppose if you have to ask....
Danny Aldham
John Wiltshire
imr...@interlog.com (Martin Imrisek) wrote:
>In message <4up9l9$h...@hobyah.cc.uq.oz.au> - j...@qits.net.au (John
>Wiltshire)Tue, 13 Aug 1996 07:07:43 GMT writes:
>:>DEC is working on a 64bit addressing system for NT AFAIK to help with
>:>the huge DB support (and to take advantage of their Alphas).
>DEC is also advertising that Linux (64 bit) exists for their Alphas!!!
I know. So?
John Wiltshire
ald...@vanbc.wimsey.com (Danny Aldham) wrote:
>John Wiltshire (j...@qits.net.au) wrote:
>: How hobbled is the perl in NT4? Anyone got any comments?
>Perl is hobbled from the point of view that you do not have access to
>any unix systems calls. So perl code that would work on Solaris, AIX,
>SCO, HP-UX, SGI, etc, etc will not run on NT. Something as simple as:
>open(MAIL,"|/usr/bin/mail") || die ; will die . The most common question
>at comp.lang.perl.misc is How do I do XXXX in NT ?
OK. I understand this. It is a fair enough comment then. How
important is this sort of thing and what is the difficulty in
converting to a native NT format (after all, NT is not Unix).
>: Why do you need telnet?
>I suppose if you have to ask....
That is exactly what I mean.
Obi Thomas
[ ... ]
> >More robust memory protection than NT.
>
> This I am interested in. Can you split the thread and explain for me?
Yes, I too would like to forget the advocacy for a moment (although it
is interesting!) and see a technical discussion of this.
Is Steve Cole referring to NT's support of 16-bit-Windows applications?
Jouko Holopainen
John Wiltshire wrote:
> You don't have to use the GUI with Win32. You can either compile a
> 'windows' application or a 'console' application - both are supported
> by the system.
Wow. This is one other thing i like about unix.
There is actually not "console" or "gui" applications, in same sense
as in nt.
> I know that NT only supports POSIX.1 and that was pretty much so the
> US government could buy it.
> What does Posix have that Win32 doesn't?
(Actually, which posix... :-)
Not much, if any. Mutexes seem to be missing, but seem to be easy to
simulate with critical sections/semaphores.
I'd rather just not write another OS hide layer again.
> >So why it [NT] is not a multiuser OS?
> >This just means that I cannot compile (or access anything
> >big) over modem.
> Well, not the kernel, but shipped with the OS I'm sure you mean! I'd
Exactly. Sorry about the mistake, it is too easy to: kernel == O.S.
(ridiculous, actually).
> rather have as many daemons (services on NT) running as a well-defined
> user, not a superuser.
And newer NT's are moving stuff (like graphics drivers) into kernel...
> This is a falling down on NT when looked at from the Unix world, but I
> personally find it not so much of a hassle (probably just I don't know
> what I'm missing?)
I cannot work from home, unless I install compiler, sources, word, excel...
(and all dll's) to my home computer. You can imagine the amount of binaries
we have on our (NT) server. Unless company buys me ATM (in my dreams...)
I do not have a long way to work (4km, 2.5miles), but a friend has
(about 90km, ~60 miles). So this is not a big hassle :-)
Starting Word over modem is impossible, but running X is just nuisance.
To emphasize MS "commitment" to Internet: a total lack of
a decent vt100 emulator (capable of handling curses, e.g. screen).
> John Wiltshire
--
Jouko Holopainen X-Net Oy
Teknologiantie 4 FIN-90570 Oulu
FINLAND GSM: +358 50 5126166
jouko.ho...@xnet.otm.fi
Steve Cole
On Sun, 18 Aug 1996, John Wiltshire wrote:
> Retail is about $800 (5 user license which is what was specified at
> the beginning).
Yikes... unlimited seats on a Solaris box for similar money. PC-NFS isn't
so cheap, however.
> Why do you need telnet?
Break a leg, work from home for a week. You need telnet.
Steve Cole
On Sun, 18 Aug 1996, John Wiltshire wrote:
> >Remote administration - could you be like me and use a box 50 miles away
> >for *everything* you do?
>
> I though you said admin, not remote use. Weren't we talking about a
> small 5 person office here?
What if you didn't have to pay travel for a consultant (maybe the
consultant sold you your accounting package), or needed a near-instant fix
from an administrator on vacation? These are possible in the UNIX world.
My bud in Dupont has to go through hell trying to do it through home,
emulating the NT interface on his machine with the 'Norton Anywhere' style
programs. He's had some good times with requestors popping up on the NT
server and locking up the emulated session when NT's wonderful
'multitasking' halted waiting for someone to hit the requestor (and thus
killing the emulation) from the physical location.
> >CDE and/or X-Windows compatibility
>
> Why is this needed for a 5 user system?
You wouldn't have to buy PCs, then - you can pick up cheap, fast,
hi-resolution Xterminals. In fact if your apps were text-based, you could
pick up VT100 terminals for free anywhere - not that I'm recommending
that. This is just one scenario, of course.
> >POSIX Compliancy
>
> Why is this needed? We have POSIX.1 (though this is c**p I know).
> Win32 does pretty much everything anyhow...
Yeah! Why are open standards required anyways? I don't see the need, do
you see a need? There must not be a need, Bill's thought of everything!
<chuckle>
> Can't beat this one. NT is getting better all the time but still has
> a way to go in maturity. When it does though, :-)
Then it'll be second best to enterprise UNIX instead of second-best to
Novell. Novell at least doesn't usually try to pretend that Netware is
equivalent, better or whatever to UNIX. Novell is often less expensive,
and there's no lack of Novell Admins, but NT *is* making progress in that
market.
> >Real-time capabilities with multi-threaded 64-bit kernel I/O (s2.5.1)
>
> This is needed for a 5 user office?
Call it an added bonus.
> >$130 (Can.) for the media of your choice if you have a SPARC-based system.
>
> media? You mean CDROM, tape etc?
Usually CDROM, yes. The current SPARC CDROMs come bundled with IPx
gateway, updates & enchancements, CDE, WABI, DiskSuite, Admin extensions,
Backup, all in the same package. My package is six CDROM discs.
$130.00 for the SPARC buys whichever version you are after - the current
SunOS 4 or SunOS 5 - your choice (unless you are running Ultra
technology).
> >More than 10,000 native applications (SPARC), way more than NT!
>
> Comes with the maturity of the system. For a small office, NT has all
> the applications you need though (I can't imagine what you'd be left
> wanting in).
So you're saying that choice isn't a positive aspect of Solaris?
> >Optimized version available for streaming video service applications.
>
> C'mon... 5 USER OFFICE! Streaming Video? MS is working on streaming
> video at the moment. I've seen demos and they have a campus wired up
> on it at the moment somewhere in the US. Doesn't even need an
> optimised version of NT.
Any major OS can do streaming video to a point, but Sun has a version
specifically for heavy usage and has had for three years. We come back to
maturity - "MS is working on streaming video at the moment". So what if
it's a 5 user office - maybe it'll be a 500 user office in two years...
they won't have to change a *thing* if they want to start tooling with the
technology... remember, Solaris is Prego.
> >Self-optimizing kernel (adds and deletes buffers as required, FI).
>
> ?? What buffers? Why?
On-the-fly performance and memory optimization with load adapting. Not
using name services much? Drop buffers. Not using TCP/IP much? Drop
bufffers. Using TCP/IP a lot? Add buffers for concurrency.
Get it?
> >More robust memory protection than NT.
>
> This I am interested in. Can you split the thread and explain for me?
UNIX has a near-uncrashable memory model which can usually only be
compromised by bugs in kernel code or a problem device.
> > >Self-optimizing UNIX file system (co-existant 8K and .5K 'block
sizes') > > Hmm. NTFS can't do this at the moment. Is it a transaction
based > system for rollback on powerdown?
Damn! Hit the justify key in Pine. Sorry for the mixed-up text from
here on... didn't feel like retyping all the stuff above :) (yes I know
Pine has an unjustify, but it didn't work... must have moved the cursor
too)
> > >Flat file system structure
(what's with this C:, D: crap?!?!) > > Well, you can just use the full
flat names if you want to. This way > is shorter (and helps windows
applications to run). How does WABI do > this?
WABI is something I don't use - I have Pentium machines all over the place
too... not much reason to burden our machines with emulation. As far as I
know, WABI emulates the C: D: mindset in its emulation.
SPARC & PPC systems, NT-like P&P on x86. > > Cool. > >
>Support for standards such as CORBA, NEO, JAVA without major changes. >
> ?? NT can't support these without _any_ changes? Look again - just >
add the DLLs.
OK, what I really meant is that these standards are already in use on
Solaris and most come bundled with, or can be added free of charge, to
Solaris. For instance, an excellent ODBC Manager comes with my S2.5 and
S2.5.1 disc sets.
> > >A very active source code resource with extensive native
support. > > I'm not sure what you mean here.
Hundreds of programs available in source code format to be compiled on
whatever architecture your UNIX runs on (well, most anyways), free of
charge. Thousands more if you look hard enough.
> > >As you say, compilers
that exploit parallel systems to their best ability. > > gcc? This is
something that is really up to the libraries and the > programmer isn't it
(unless you are looking at a higher level language > than C)?
I'm using GCC but the Sun compiler makes better code, I'm told. There's
something going on there if Sun's compiler is 'better'. Regardless,
programs compiled on Solaris are 'forced' into behaving, to a degree.
>Extremely good scaling support for SMP machines - proven! > > Again,
depends on the application. NT hasn't been tested well enough > on more
than about 8 CPUs for good results. The x86 architecture > kinda breaks
down the performance here.
So you're saying what? It doesn't depend at all, Solaris systems are
proven to work very well in SMP configurations.
> > >Mature OS means that security flaws have
been discovered and documented. > > NT was designed from the ground up
for security. _Very few_ flaws > have been discovered in the system and
all those that have are well > documented (or there would be more
specifics in the usenet wars). I > am on just about every NT security
mailing list and I still see more > flaws in Solaris than NT coming
through for all your 'mature OS' > claims.
So imagine what you *haven't* heard about NT's security flaws yet. Wow,
eh? I can get any file I want off one of my customer's NT boxes on the
other end of a leased line, without him knowing it. And I can write to
the machine too, as Admin... without custom security software.
As before, I won't release the details... sorry.
> > >Proven reliability in
high-availability and high-stress conditions. > > Since when was the 5
user system high-stress? There are times when I > will not advocate NT,
but for a small 5 user office I think it is a > good idea.
But we come back to the "what if we grow" scenario. There are so many
SPARC systems out there that, with the addition of another 128MB of cheap
memory, are still running fairly large applications on single processors.
You have to remember, we've had Pentium and Pentium Pro level speeds for
years, so these machines have provided *excellent* cost effective service
for years. Solaris is in the same boat - want Internet connectivity, real
Sendmail, GNU utilities, remote administration, etc. etc.? It'll be there
when you're ready for it... no software to buy, no upgrades to do.
I've already proven that Solaris can be less expensive than NT, so what's
the beef? It can't be cost-effectiveness.
Like you, I think NT has its place, but I think more important to the
companies should be experienced, seasoned administrators who can do much
more than just 'make it work'.
Alan Osborne
Technical pros and cons in all these discussions are very interesting,
but the bottom line is that a company will buy the systems that enable
the sacking of the most people. Perhaps we need a new, more relevant
benchmark.....
--
AlanO
Edmond
Danny Aldham wrote:
>
> John Wiltshire (j...@qits.net.au) wrote:
> : How hobbled is the perl in NT4? Anyone got any comments?
>
> Perl is hobbled from the point of view that you do not have access to
> any unix systems calls. So perl code that would work on Solaris, AIX,
> SCO, HP-UX, SGI, etc, etc will not run on NT. Something as simple as:
> open(MAIL,"|/usr/bin/mail") || die ; will die . The most common question
> at comp.lang.perl.misc is How do I do XXXX in NT ?
There are mail binaries around. You simply have to have a pathname for
/usr/bin/mail. This is quite trivial. Hell, I've hacked one together
for CGI scripting before. Worked well too. I don't think it's fair to
say "any" system programs either. There are many GNU and POSIX level
implementations for NT.
--
Edmond Underwood
U Software
Bench32 1.20 - The new threshhold in system benchmarking
Bench32 web site: http://www.rmii.com/~underwoe/bench32.html
Jack Love
On Sat, 17 Aug 1996 01:05:22 GMT, ev...@telly.org (Evan Leibovitch) wrote:
>In article <4v0o45$o...@knot.queensu.ca>,
> Jianlin Chang <ch...@mjs1.phy.queensu.ca> wrote:
>
>>It is hard to say, DEC UNIX is not performing well, it's not certain
>>how well they will perform with Alpha NT. I can't see DEC's strong
>>commitment to UNIX.
>
Digital's historical personality is to 'support' MANY different
operating systems on their hardware. This was a strength
AND a weakness as their software support often left much
to be desired.
Edmond Underwood
On Mon, 19 Aug 1996, Phil Brown wrote:
> In article <3210FA...@Colorado.Edu> you write:
> >over 6 GB large). There is also no one UNIX. BSDi, NeXTStep, etc. don't
> >scale at all. Solaris also doesn't scale linearly at all on an x86 dual
> >P100.
>
> limitations of the hardware, not the OS, I've heard.
> --
>
>
Perhaps so, but your comparing apples to apples now. I don't think the
x86 version has the exact same kernel anyway (the key software piece in
scalability).
-Edmond
Wayne Hyde
John Wiltshire <j...@qits.net.au> wrote in article
<4v76qd$q...@hobyah.cc.uq.oz.au>...
> >: It is hard to say, DEC UNIX is not performing well, it's not certain
> >: how well they will perform with Alpha NT. I can't see DEC's strong
> >: commitment to UNIX.
> >how NT doesn't scale beyond 4 or 5 Pentuims. I think it would be a
> >screamer on 4 or 5 Alpha's.
> NT4 scales well to about 8 CPUs now.
I've seen a few tests that show NT (even 3.51) scale well up to 4 CPUs, with
NT4 up to about 8 CPUs. I have *not* seen a test for above 8 CPUs. Someone
reported that PCWEEK didn't test Sequent's 32CPU NT box or any other high-end
hardware. Does anyone have a URL or case study of NT's scalability on
high-end boxes (>8 CPU)?
NT's scalability also depends highly on the system architecture. On my
el-cheapo Tyan Tomcat II motherboard with 256k shared cache and dual-P90's, I
get about 70% scalability. NT should scale much better on a high-end board
with a C-BUS, separate caches for each CPU, etc. I think I'll live with my
$500 SMP board (cost includes CPUs, cache, and board).
--
Wayne Hyde | System Administrator
w...@cis.ufl.edu | Fla. Cooperative Fish & Wildlife Research Unit
http://www.cis.ufl.edu/~wjh | and the
I speak for me, not them -> | U of Florida Department of Surveying and Mapping
Jon A. Maxwell
Edmond wrote:
] Do you honestly think programming widgets is easier than
] Windows? Just curious.
I _know_ this: programming wxWindows (C++ object layer over
Motif, Xt, or Olit) is about the easiest thing imaginable.
Far easier than "Windows". And it runs faster than OLE!
] > numbers for a Cray (now Sun) SuperServer going from 24 to 48 processors
] > (41967 to 82522 - a ratio of 1.97).
]
] to be true. This looks much like Digital's benchmark showing how their
] 64-bit CPU makes databses 8x faster than 32-bit CPU's (the data was well
] over 6 GB large).
If you were concerned at all about the database being faster,
your database must be pretty big to start with (like 6 GB).
--
thur Mail Address: LordA...@vt.edu or jmax...@vt.edu
n r
a JAMax One Ring to rule them all, One Ring to find them,
h o w One Ring to bring them all and in the darkness bind them
tan lle --Tolkien (The Lord Of The Rings)
John Wiltshire
Steve Cole <co...@post.kosone.com> wrote:
>On Sun, 18 Aug 1996, John Wiltshire wrote:
>> >Remote administration - could you be like me and use a box 50 miles away
>> >for *everything* you do?
>>
>> I though you said admin, not remote use. Weren't we talking about a
>> small 5 person office here?
>What if you didn't have to pay travel for a consultant (maybe the
>consultant sold you your accounting package), or needed a near-instant fix
>from an administrator on vacation? These are possible in the UNIX world.
>My bud in Dupont has to go through hell trying to do it through home,
>emulating the NT interface on his machine with the 'Norton Anywhere' style
>programs. He's had some good times with requestors popping up on the NT
>server and locking up the emulated session when NT's wonderful
>'multitasking' halted waiting for someone to hit the requestor (and thus
>killing the emulation) from the physical location.
This, again, is interesting. All of the network admin tools that are
shipped with NT and the BackOffice programs have RPC interfaces to the
'invisible' services which enable seamless remote admin. There really
is no need to use the 'PC Anywhere' style remote applications unless
you really have to support old Win3.x style applications (which is
just as much of a headache on Unix).
>> >CDE and/or X-Windows compatibility
>>
>> Why is this needed for a 5 user system?
>You wouldn't have to buy PCs, then - you can pick up cheap, fast,
>hi-resolution Xterminals. In fact if your apps were text-based, you could
>pick up VT100 terminals for free anywhere - not that I'm recommending
>that. This is just one scenario, of course.
If that's what you want, I guess it works. I'm more inclined to go
for the distribution of PC power and using the least CPU on the server
for a 5 user situation. Counting the workstations, what sort of
hardware bill are you up for using Solaris for this 5 user system (no
breakdown necessary).
>> >POSIX Compliancy
>>
>> Why is this needed? We have POSIX.1 (though this is c**p I know).
>> Win32 does pretty much everything anyhow...
>Yeah! Why are open standards required anyways? I don't see the need, do
>you see a need? There must not be a need, Bill's thought of everything!
Dave actually (y'know - PDP11, VMS etc.). Bill kept out of this one,
thank goodness!
Why does everyone assume that people that appreciate NT think BG is a
wonderful person who spouts wisdom?
>> Can't beat this one. NT is getting better all the time but still has
>> a way to go in maturity. When it does though, :-)
>Then it'll be second best to enterprise UNIX instead of second-best to
>Novell. Novell at least doesn't usually try to pretend that Netware is
>equivalent, better or whatever to UNIX. Novell is often less expensive,
>and there's no lack of Novell Admins, but NT *is* making progress in that
>market.
Didn't Novell just ditch Unixware to a company that could actually do
something useful with it? NT and Novell have been head to head for a
while now and while NT's sales are still increasing, Novell has been
pretty stagnant. I can see why Novell Admins are easy to find - just
go to Social Security. ;-)
>> >$130 (Can.) for the media of your choice if you have a SPARC-based system.
>>
>> media? You mean CDROM, tape etc?
>Usually CDROM, yes. The current SPARC CDROMs come bundled with IPx
>gateway, updates & enchancements, CDE, WABI, DiskSuite, Admin extensions,
>Backup, all in the same package. My package is six CDROM discs.
>$130.00 for the SPARC buys whichever version you are after - the current
>SunOS 4 or SunOS 5 - your choice (unless you are running Ultra
>technology).
Why the extra cost then if you are spending money for the OS
originally? I already get NT on CDROM without spending money? I
don't think I've got the point here...
>> >More than 10,000 native applications (SPARC), way more than NT!
>>
>> Comes with the maturity of the system. For a small office, NT has all
>> the applications you need though (I can't imagine what you'd be left
>> wanting in).
>So you're saying that choice isn't a positive aspect of Solaris?
Not at all. It is definitely positive, but not a _big_ positive.
There are millions of native Win16 applications but I wouldn't use it.
>> >Optimized version available for streaming video service applications.
>>
>> C'mon... 5 USER OFFICE! Streaming Video? MS is working on streaming
>> video at the moment. I've seen demos and they have a campus wired up
>> on it at the moment somewhere in the US. Doesn't even need an
>> optimised version of NT.
>Any major OS can do streaming video to a point, but Sun has a version
>specifically for heavy usage and has had for three years. We come back to
>maturity - "MS is working on streaming video at the moment". So what if
>it's a 5 user office - maybe it'll be a 500 user office in two years...
>they won't have to change a *thing* if they want to start tooling with the
>technology... remember, Solaris is Prego.
OK. I know MS has been in serious talks with pay TV providers over
here for the use of NT for Video on Demand services to the general
population. As these telcos are fairly heavily Unix based, I was
suprised (pleasently) that they were even considering NT.
>> >Self-optimizing kernel (adds and deletes buffers as required, FI).
>>
>> ?? What buffers? Why?
>On-the-fly performance and memory optimization with load adapting. Not
>using name services much? Drop buffers. Not using TCP/IP much? Drop
>bufffers. Using TCP/IP a lot? Add buffers for concurrency.
..haven't looked at this in NT. Not qualified enough to reply.
Sorry.
>> >More robust memory protection than NT.
>>
>> This I am interested in. Can you split the thread and explain for me?
>UNIX has a near-uncrashable memory model which can usually only be
>compromised by bugs in kernel code or a problem device.
So does NT. Unless Unix has gone the Mach way lately I'm not really
impressed here.
>> > >Self-optimizing UNIX file system (co-existant 8K and .5K 'block
>sizes') > > Hmm. NTFS can't do this at the moment. Is it a transaction
>based > system for rollback on powerdown?
>SPARC & PPC systems, NT-like P&P on x86. > > Cool. > >
>>Support for standards such as CORBA, NEO, JAVA without major changes. >
>> ?? NT can't support these without _any_ changes? Look again - just >
>add the DLLs.
>OK, what I really meant is that these standards are already in use on
>Solaris and most come bundled with, or can be added free of charge, to
>Solaris. For instance, an excellent ODBC Manager comes with my S2.5 and
>S2.5.1 disc sets.
I get an ODBC manager with NT too, Java for free (no major changes
either), DCOM for free, don't know what NEO is...
>> > >A very active source code resource with extensive native
>support. > > I'm not sure what you mean here.
>Hundreds of programs available in source code format to be compiled on
>whatever architecture your UNIX runs on (well, most anyways), free of
>charge. Thousands more if you look hard enough.
I agree that these need porting to NT, but most major developers seem
to be shying away from providing source code to clients and build the
executables themselves. NT's POSIX compliance should be able to
handle most of these style of applications (though a 3rd party
X-Windows manager may be required). For something that runs well on
the other hand, it will need redesign and I take your point.
[SMP -]
>So you're saying what? It doesn't depend at all, Solaris systems are
>proven to work very well in SMP configurations.
Of course it depends, but from what I've seen, NT scales very well up
to 4 CPUs with 3.51 and about 8 for 4.0. I know that Solaris goes to
64 or so, but as it was said somewhere else, how many 8+ CPU machine
do you know?
>So imagine what you *haven't* heard about NT's security flaws yet. Wow,
>eh? I can get any file I want off one of my customer's NT boxes on the
>other end of a leased line, without him knowing it. And I can write to
>the machine too, as Admin... without custom security software.
Sounds like he set it up badly.
Most of the security holes in Unix are vectored through suid programs.
NT does not offer the possibility of using suid and so closes this
hole. Exploitations have been published for IIS (by 3rd parties and
confirmed by MS) but I haven't seen any true flaws that don't involve
admin doing something stupid.
You can't make the assertion that because there have been few flaws
found that there must be vast quantities out there. I'm not blind
enough to assume that there are no flaws, but given the track record
so far I'd say the number is far fewer then the original Unixes.
Don't get me wrong though - I don't mean that Solaris/SunOS is not
secure. I'm just pointing out that NT isn't either.
>But we come back to the "what if we grow" scenario. There are so many
>SPARC systems out there that, with the addition of another 128MB of cheap
>memory, are still running fairly large applications on single processors.
>You have to remember, we've had Pentium and Pentium Pro level speeds for
>years, so these machines have provided *excellent* cost effective service
>for years. Solaris is in the same boat - want Internet connectivity, real
>Sendmail, GNU utilities, remote administration, etc. etc.? It'll be there
>when you're ready for it... no software to buy, no upgrades to do.
Why is sendmail such a plus. Many secure systems offer the selling
point that their product is _not_ sendmail based.
NT - internet connectivity, GNU utilities (like who cares anyway but
they exist), remote admin etc. etc as well as far superior integration
with existing DOS/Windows machines (which are the most common systems
out there for which people are buying new servers).
>I've already proven that Solaris can be less expensive than NT, so what's
>the beef? It can't be cost-effectiveness.
You did? I though that was what started this thread. You put forward
a hypothesis and I matched it.
I though it went that NT can also be less expensive than Solaris.
>Like you, I think NT has its place, but I think more important to the
>companies should be experienced, seasoned administrators who can do much
>more than just 'make it work'.
True. Can't agree more. This is why I enjoy knowing how to
administer NT (easier work for more pay).
John Wiltshire
Jouko Holopainen <jouko.ho...@xnet.otm.fi> wrote:
>John Wiltshire wrote:
>> You don't have to use the GUI with Win32. You can either compile a
>> 'windows' application or a 'console' application - both are supported
>> by the system.
>Wow. This is one other thing i like about unix.
>There is actually not "console" or "gui" applications, in same sense
>as in nt.
What do you mean? A 'console' application can spawn itself as many
traditional 'windows' as it likes and a 'windows' application can
spawn consoles? The compilation just depends on whether your programs
entry point is main() or WinMain() and what parameters the OS passes
to it on startup (eg. Window state etc.)
>> I know that NT only supports POSIX.1 and that was pretty much so the
>> US government could buy it.
>> What does Posix have that Win32 doesn't?
>(Actually, which posix... :-)
>
>Not much, if any. Mutexes seem to be missing, but seem to be easy to
>simulate with critical sections/semaphores.
Nope - CreateMutex() exists quite happily in Win32, as well as
critical sections and semaphores.
>I'd rather just not write another OS hide layer again.
Nor me. The sooner people see that Win32 is the way to the future the
better! (joke).
We are using Win32 for everything from real-time embedded systems to
our horsepower DB systems. One code base sure is nice!
>> >So why it [NT] is not a multiuser OS?
>> >This just means that I cannot compile (or access anything
>> >big) over modem.
>> Well, not the kernel, but shipped with the OS I'm sure you mean! I'd
>Exactly. Sorry about the mistake, it is too easy to: kernel == O.S.
>(ridiculous, actually).
>> rather have as many daemons (services on NT) running as a well-defined
>> user, not a superuser.
>And newer NT's are moving stuff (like graphics drivers) into kernel...
Graphics drivers used to exist as 2 parts - the kernel mode part and
the user mode part. Now they are entirely kernel mode.
Doesn't Unix use kernel mode video drivers? (well most of them)
>> This is a falling down on NT when looked at from the Unix world, but I
>> personally find it not so much of a hassle (probably just I don't know
>> what I'm missing?)
>I cannot work from home, unless I install compiler, sources, word, excel...
>(and all dll's) to my home computer. You can imagine the amount of binaries
>we have on our (NT) server. Unless company buys me ATM (in my dreams...)
>I do not have a long way to work (4km, 2.5miles), but a friend has
>(about 90km, ~60 miles). So this is not a big hassle :-)
BTW - I'm in Australia so I don't use Imperial units. :-)
>Starting Word over modem is impossible, but running X is just nuisance.
I know what you mean here. DCOM may fix this (but I'm not holding my
breath).
I have no problems installing Office, compiler (dll's come with the
install strangely enough). I do develop for NT so I do know the size
of the binaries...
Source code is fine over a modem using SourceSafe to check them out to
your local drive.
Copyright law allows you to install software on your home computer if
you use it at work (in Australia).
>To emphasize MS "commitment" to Internet: a total lack of
>a decent vt100 emulator (capable of handling curses, e.g. screen).
I don't think that this is right up there on their list of priorities
somehow? There's much bigger things to fix, like a good directory
service...
John Wiltshire
Steve Cole <co...@post.kosone.com> wrote:
>On Sun, 18 Aug 1996, John Wiltshire wrote:
>> Retail is about $800 (5 user license which is what was specified at
>> the beginning).
>Yikes... unlimited seats on a Solaris box for similar money. PC-NFS isn't
>so cheap, however.
>> Why do you need telnet?
>Break a leg, work from home for a week. You need telnet.
Again, why? I dial in and use RAS. Not a real difficulty here.
Edmond
Steve Cole wrote:
>
> On Tue, 20 Aug 1996, John Wiltshire wrote:
>
> > Doesn't Unix use kernel mode video drivers? (well most of them)
>
> the drivers in as kernel modules, which can be dropped and can usually be
> kept from crashing the system under failure. However, you have to
If you video driver crashes under Solaris, your stuck pal. One can keep the
rest of the kernel from falling also (same as in NT any version), but you lose
control of your GDI. As far as I know, there is no mechanism for restarting a
driver on the fly by the Solaris kernel itself (same as NT any version). A
video driver in the kernel doesn't really change anything from NT 3.51 to 4.0.
A crashed driver froze the GDI since nothing could be written or read from the
port (int NT 3.51). The former isn't technically a crash, but you would still
have to restart the system if you wanted control back. The question is: how
often will a driver crash? Video driver failures aren't common. Most UNIX
OS's load all drivers at the kernel level. According to your logic, they would
be that many times less stable than NT (since video drivers are the only kernel
mode drivers under NT). Obviously, I don't agree with your logic.
> remember that in the workstation market, anyway, the hardware choices are
> 1/100th of the choices in the PC market (a *very* good thing) because most
> of the hardware was pretty good ten years ago. In short, the
> manufacturers are intimate with their own hardware and can make
> bulletproof, optimized code for them.
>
What about Intergraph, HP (uses the same parts as the 8000 in the new XA
models), IBM, ALR, STG, and the other workstation vendors?
> NT has none of these advantages and has to deal with YGWYPF hardware all
> the time. Unfortunately, so does Solarisx86, and it pays to stay with
> recommended hardware.
>
This last statement is incredibly wrong. Solaris has versions for PPC (coming
up), SPARC/Ultra, and x86. It runs on almost as much hardware as NT. It's
problems are NT's problems also. There are also very good workstation vendors
that sell NT boxes custom made with good hardware. I haven't heard a great
deal of problems from those vendors.
Steve Cole
On Tue, 20 Aug 1996, John Wiltshire wrote:
> Doesn't Unix use kernel mode video drivers? (well most of them)
Someone may correct me if I'm wrong, but as far as I'm aware Solaris loads
the drivers in as kernel modules, which can be dropped and can usually be
kept from crashing the system under failure. However, you have to
remember that in the workstation market, anyway, the hardware choices are
1/100th of the choices in the PC market (a *very* good thing) because most
of the hardware was pretty good ten years ago. In short, the
manufacturers are intimate with their own hardware and can make
bulletproof, optimized code for them.
NT has none of these advantages and has to deal with YGWYPF hardware all
the time. Unfortunately, so does Solarisx86, and it pays to stay with
recommended hardware.
Cheers,
Steve Cole
On Tue, 20 Aug 1996, John Wiltshire wrote:
> This, again, is interesting. All of the network admin tools that are
> shipped with NT and the BackOffice programs have RPC interfaces to the
> 'invisible' services which enable seamless remote admin. There really
> is no need to use the 'PC Anywhere' style remote applications unless
> you really have to support old Win3.x style applications (which is
> just as much of a headache on Unix).
Describe what you can do with the RPC services in these new, improved
products. Can you work with a database remotely, for instance? Can you
mount, format, use and unmount a hard drive from remote? Can you, in
fact, do absolutely anything over a modem that you can do from the console
(barring the physical addition of new equipment, of course), and do it in
a way that is more or less identical to what you expect at the console?
In Solaris, I can, unless a program were specifically designed to require
a user at the console (which I am not aware of at this time).
> If that's what you want, I guess it works. I'm more inclined to go
> for the distribution of PC power and using the least CPU on the server
> for a 5 user situation. Counting the workstations, what sort of
> hardware bill are you up for using Solaris for this 5 user system (no
> breakdown necessary).
$1200 per Xterminal, $950 for an extra 64MB SIMM for the server to handle
the extra load. So, I'd recommend an Ultra 1 server which can grow a
great deal, five Xterminals and 128MB memory. All for about $24,000
Canadian which includes a extended package for the Ultra (20" monitor,
2.1G drive, 64MB option). Notice I didn't price this out to the penny.
Throw in cabling, hub and you have SPARC workstation power on five desks,
a system with headroom, Internet style messaging and compatibility, file
sharing, and all the other things that go with Solaris that I mentioned
earlier. You could probably scale a system like that up to around 30
users running light CAD, or much more running small databases, accounting,
etc.
So, can you do it for that in an NT environment, with the headroom,
performance and Internet compatibility (total, mind, no major restrictions
or incompatibilities)? Here's the price tag (rough) for a 30 user network
built on the model above, which includes extra RAM, etc:
$65,000 - 512MB RAM, 16GB (Metadisk) storage, 30 Xterminals
Double the power with an Ultra2 server, add $11,000 (rough)
NT solution (my estimate):
$141,000 - 200MB RAM, Dual Pentium 166, 16GB storage, 30 Pentium 120s
Scary, huh? Hell, you can add another two Ultra2 servers and a lot more
Xterminals for that kind of cash.
> Why does everyone assume that people that appreciate NT think BG is a
> wonderful person who spouts wisdom?
It goes with the territory, I guess.
> Didn't Novell just ditch Unixware to a company that could actually do
> something useful with it? NT and Novell have been head to head for a
> while now and while NT's sales are still increasing, Novell has been
> pretty stagnant. I can see why Novell Admins are easy to find - just
> go to Social Security. ;-)
I agree with you, but I'm not at all convinced that NT deserves its
victories in most cases. Novell didn't know what to do with UNIXware.
> Why the extra cost then if you are spending money for the OS
> originally? I already get NT on CDROM without spending money? I
> don't think I've got the point here...
The license for the software comes with the machine. You then pay for the
media (manuals and CDs) for whatever flavour of OS you want (SunOS 4 or
SunOS 5).
> OK. I know MS has been in serious talks with pay TV providers over
> here for the use of NT for Video on Demand services to the general
> population. As these telcos are fairly heavily Unix based, I was
> suprised (pleasently) that they were even considering NT.
Phah. Sun has sold loads of their Enterprise servers to telephone
companies and video houses in the last year. There's nothing in an NT
solution that can touch that price/performance/uptime blend. In fact, RMC
here in Kingston just bought an Enterprise 4000 because when they compared
pricing and solutions to DEC, HP, SGI and NT, everyone but DEC asked RMC
to lower the spec because they couldn't touch it, and DEC wanted about
double the price for a cluster technology. The NT guys were speechless...
"You mean just about everything can fail and the machine can still run?"
Sun's had a video-server specific version of Solaris for years, too.
I don't think ANYONE is losing any breath at Sun over this one.
> So does NT. Unless Unix has gone the Mach way lately I'm not really
> impressed here.
Solaris is very Mach-like. It's a tiny microkernel which is built upon
and torn down in loadable modules as the system runs. That's right, the
kernel will even drop parts of itself out of memory if those things are
not being used. Try this on Solaris: type "ps -aex" on a machine that has
been running unattended for days with no users logged in. It will take
longer while the kernel adds support for the command back in. Run "ps
-aex" again... BOOM, printout. This is the way it's been explained to me
at any rate - I am *not* a kernel programmer, I just listen well :)
In fact, Sun's still improving - they've copied a number of ideas from a
test operating system they called Spring, which has greatly improved the
concepts in Solaris and will continue to do so as they build Doors and
other new ideas into Solaris.
> Hmm. NTFS can't do this at the moment. Is it a transaction
> based system for rollback on powerdown?
Up to you. You can choose either method.
> I get an ODBC manager with NT too, Java for free (no major changes
> either), DCOM for free, don't know what NEO is...
Is it an ODBC manager? NEO is NEO - it's not an acronym :)
> to 4 CPUs with 3.51 and about 8 for 4.0. I know that Solaris goes to
> 64 or so, but as it was said somewhere else, how many 8+ CPU machine
> do you know?
I can put my hand on three of them that were sold in the last month in the
little burgh of Kingston. These things will be commonplace with pricing
like 1/2 million installed and premier service contracted for a 14-way,
redundant, 1.5GB monster mainframe-eater. And that pricing just dropped
$23,000 in memory discounts alone this week. I can also put my hands on
five Compaq 4500 and 5000 series dual-Pentium machines running SCO, NT and
Solaris mix. Multi-processor will be commonplace in two years.
> Sounds like he set it up badly.
Probably. It doesn't change the fact that it was *easy* and I'm no NT
guru!
> Most of the security holes in Unix are vectored through suid programs.
> NT does not offer the possibility of using suid and so closes this
> hole. Exploitations have been published for IIS (by 3rd parties and
> confirmed by MS) but I haven't seen any true flaws that don't involve
> admin doing something stupid.
The underground on the 'net doesn't share your enthusiasm for NT's
security <grin> Not that I'm plugged into the deeper levels (anymore),
but the entire source tree for NT is out there for FTP if you know where
to get it - the hackers know *all about* NT and its possible problems, the
same as Solaris.
> You can't make the assertion that because there have been few flaws
> found that there must be vast quantities out there. I'm not blind
> enough to assume that there are no flaws, but given the track record
> so far I'd say the number is far fewer then the original Unixes.
No question. NT isn't as capable, and therefore isn't as problematical.
I won't argue with you on this one. Security is a very big, dirty, nasty
issue with the Internet no matter what you have for machinery and OS.
Speaking of security, the 'net isn't your only problem. One of our local
'net companies just lost all their machinery last night, including a Sun
Ultra 170 Creator. Crying time... I'm backing up my systems with tapes to
replace the current ones in security deposit, just to be safe. While I
have a motion sensor trained directly on the hardware, I don't like the
thought of walking into an empty office tomorrow.
> Don't get me wrong though - I don't mean that Solaris/SunOS is not
> secure. I'm just pointing out that NT isn't either.
Ah. Should've read this line first. :)
> Why is sendmail such a plus. Many secure systems offer the selling
> point that their product is _not_ sendmail based.
Sendmail *can* be a security risk, and *can* be painful to operate to the
neophyte to mid-level admin. Sendmail is so incredibly powerful, however,
and is 100% compatible with the Internet. Sendmail is really a batch
programming and execution language as far as I'm concerned.
> NT - internet connectivity, GNU utilities (like who cares anyway but
> they exist), remote admin etc. etc as well as far superior integration
> with existing DOS/Windows machines (which are the most common systems
> out there for which people are buying new servers).
Why superior? For $0 for unlimited seats, I can have my Solaris box look
*exactly* like an NT server to Windows clients, with the exception that I
don't have (and wouldn't use if I did) WINS.
> You did? I though that was what started this thread. You put forward
> a hypothesis and I matched it.
No, you didn't. You didn't come up with a similar price at all, based on
initial cost OR cost to grow to thirty machines (in this message). Dare I
say that NT would cost you a *great* deal more, considering that
Xterminals are just plugged into the network and turned on, while
estimates of PC maintenance range from $3000 to $5000 over a three year
period. If an Xterminal dies, you unplug it, throw it away, plug in the
new one and turn it on.
> I though it went that NT can also be less expensive than Solaris.
It can be less expensive, but not when Solaris is let loose to work with
its own standards instead of wrapping around the MS mindset. And it can
be less expensive there if you want to grow your network and connect it to
the Internet, etc. And just think of how well those new Network Computers
will work with Solaris... did you know that Sun has sold millions of them
and hasn't shipped a finished product yet?
> True. Can't agree more. This is why I enjoy knowing how to
> administer NT (easier work for more pay).
What do you make? I know UNIX admins that answer email all day making
$90,000+/a which work about four weeks out of the year upgrading,
updating and improving facilities. Most of their job is handing out and
designing network segments... after all, their machines don't give them
much grief. Hell, what do you think *I* make? :)
And *anyone* can say their an NT administrator, but not many can say with
any confidence that they even understand the concepts behind Sendmail.
Queen's University estimates that it would cost them three million in
mistakes and wrong decisions if they were to replace their current
MIT... and they bought about 300 Sun machines this year I'm told... now
*that* is job security.
Danny Aldham
X-Newsreader: TIN [version 1.2 PL2]
John Wiltshire (j...@qits.net.au) wrote:
: Steve Cole <co...@post.kosone.com> wrote:
: >On Sun, 18 Aug 1996, John Wiltshire wrote:
: >> Why do you need telnet?
: >Break a leg, work from home for a week. You need telnet.
: Again, why? I dial in and use RAS. Not a real difficulty here.
Since NT does not include a command line or shell, telnet to NT doesn't
give you much. Telnet without a $prompt would be like working with only
1 hand tied behind your back. But I understand that a shell _is_ included
with NT4.0 , so to get to it (from anything beside another MS product),
telnet is needed. Why MicroSoft don't include a telnetd is beyond me. Probably
thought up by the same people that want to hobble NT workstation by only
allowing 10 incoming HTTP requests per hour.
Danny Aldham
Bryan Althaus
Edmond (unde...@Colorado.Edu) wrote:
: Steve Cole wrote:
Edmond (unde...@Colorado.Edu) wrote:
: Steve Cole wrote:
Edmond (unde...@Colorado.Edu) wrote:
: Steve Cole wrote:
Guys, I think it's time for email.... :-)
Edmond
Steve Cole wrote:
>
> On Tue, 20 Aug 1996, John Wiltshire wrote:
>
> > If that's what you want, I guess it works. I'm more inclined to go
> > for the distribution of PC power and using the least CPU on the server
> > for a 5 user situation. Counting the workstations, what sort of
> > hardware bill are you up for using Solaris for this 5 user system (no
> > breakdown necessary).
>
> $1200 per Xterminal, $950 for an extra 64MB SIMM for the server to handle
> the extra load. So, I'd recommend an Ultra 1 server which can grow a
> great deal, five Xterminals and 128MB memory. All for about $24,000
> Canadian which includes a extended package for the Ultra (20" monitor,
> 2.1G drive, 64MB option). Notice I didn't price this out to the penny.
I just looked at Sun pricing. $24,000 for an Ultra I seems low unless your
in academia. What is your package? We have X-terms with 17" monitors. They
run from $1950 to $2200 at academic prices.
> Throw in cabling, hub and you have SPARC workstation power on five desks,
> a system with headroom, Internet style messaging and compatibility, file
> sharing, and all the other things that go with Solaris that I mentioned
> earlier. You could probably scale a system like that up to around 30
> users running light CAD, or much more running small databases, accounting,
> etc.
>
> So, can you do it for that in an NT environment, with the headroom,
> performance and Internet compatibility (total, mind, no major restrictions
> or incompatibilities)? Here's the price tag (rough) for a 30 user network
> built on the model above, which includes extra RAM, etc:
>
Yes.
> $65,000 - 512MB RAM, 16GB (Metadisk) storage, 30 Xterminals
>
> Double the power with an Ultra2 server, add $11,000 (rough)
>
Ultra II, about $54,000 with 128 MB ram.
> NT solution (my estimate):
>
> $141,000 - 200MB RAM, Dual Pentium 166, 16GB storage, 30 Pentium 120s
>
> Scary, huh? Hell, you can add another two Ultra2 servers and a lot more
> Xterminals for that kind of cash.
>
We have X-terminals also. They can serve as X-terms or NT terms. There's an
option at setup for it. You need to have N-Trigue running on your NT Server.
It's about $3000-$10000 depending on the number of clients. A 4 CPU Pentium
PRO from STB (without monitor but with 64 MB ECC Ram 4-way interleved, 4 Fast
Ethernet cards, A 9 GB Ultra SCSI controller, Video card with 2 MB VRAM, and
4 Pentium PRO CPU's went for $9500). The prices of the two setups would be
much close than your putting it at. An X-term couldn't perform up to par
with desktops anyway. That's one of the reasons that their not used
everywhere here. Their no good for GIS software, and running NFS is nothing
like having your local disk spinning. Even with DiskSuite over Fast/Wide
chains.
> The license for the software comes with the machine. You then pay for the
> media (manuals and CDs) for whatever flavour of OS you want (SunOS 4 or
> SunOS 5).
>
There's also licensing for Multiuser setups (there's restrictions on users
and nfs/etc. in the client Solaris license), compilers, libraries, etc. It
adds up fast.
> > So does NT. Unless Unix has gone the Mach way lately I'm not really
> > impressed here.
>
> Solaris is very Mach-like. It's a tiny microkernel which is built upon
> and torn down in loadable modules as the system runs. That's right, the
> kernel will even drop parts of itself out of memory if those things are
> not being used. Try this on Solaris: type "ps -aex" on a machine that has
> been running unattended for days with no users logged in. It will take
> longer while the kernel adds support for the command back in. Run "ps
> -aex" again... BOOM, printout. This is the way it's been explained to me
> at any rate - I am *not* a kernel programmer, I just listen well :)
>
I think someone explained it oddly. It seems to me that those software
portions have been paged out.
. You can choose either method.
> > Most of the security holes in Unix are vectored through suid programs.
> > NT does not offer the possibility of using suid and so closes this
> > hole. Exploitations have been published for IIS (by 3rd parties and
> > confirmed by MS) but I haven't seen any true flaws that don't involve
> > admin doing something stupid.
>
> The underground on the 'net doesn't share your enthusiasm for NT's
> security <grin> Not that I'm plugged into the deeper levels (anymore),
> but the entire source tree for NT is out there for FTP if you know where
> to get it - the hackers know *all about* NT and its possible problems, the
> same as Solaris.
>
If you look at NT 4.0, some security holes were patched up. Everything has
it's holes.
> > NT - internet connectivity, GNU utilities (like who cares anyway but
> > they exist), remote admin etc. etc as well as far superior integration
> > with existing DOS/Windows machines (which are the most common systems
> > out there for which people are buying new servers).
>
> Why superior? For $0 for unlimited seats, I can have my Solaris box look
> *exactly* like an NT server to Windows clients, with the exception that I
> don't have (and wouldn't use if I did) WINS.
>
Solaris supports NT's peer networking model and it's domain model? How does
it handle long file names for DOS? I think NetBEUI support is better on the
NT side. This is what WFWG wants to do.
> No, you didn't. You didn't come up with a similar price at all, based on
> initial cost OR cost to grow to thirty machines (in this message). Dare I
> say that NT would cost you a *great* deal more, considering that
> Xterminals are just plugged into the network and turned on, while
> estimates of PC maintenance range from $3000 to $5000 over a three year
> period. If an Xterminal dies, you unplug it, throw it away, plug in the
> new one and turn it on.
>
PC's have 3-5 year warranties now. If something breaks, you get new parts
from the vendor. Maintenance bills for UNIX workstations can get high too.
University deals generally save you 75% off of the regular price.
> It can be less expensive, but not when Solaris is let loose to work with
> its own standards instead of wrapping around the MS mindset. And it can
> be less expensive there if you want to grow your network and connect it to
> the Internet, etc. And just think of how well those new Network Computers
> will work with Solaris... did you know that Sun has sold millions of them
> and hasn't shipped a finished product yet?
>
Same can be said of NT. Again, you can run terminals now from Tek, Wyse,
HDS, and others for NT AND UNIX on the same box.
> What do you make? I know UNIX admins that answer email all day making
> $90,000+/a which work about four weeks out of the year upgrading,
> updating and improving facilities. Most of their job is handing out and
> designing network segments... after all, their machines don't give them
> much grief. Hell, what do you think *I* make? :)
>
I imagine that UNIX admins make more than NT admins due to the simplicity of
the NT networks. I do "think" (I don't really know and don't want to offend
anyone) that UNIX admins are generally more knowledgeable about networking
concepts in general as well as network design, etc. This is typically
because it's what is taught at the University level; so, one would get
hammered with more theory there than at some NT Admin class.
Bryan Althaus
: >
: > Double the power with an Ultra2 server, add $11,000 (rough)
: >
: Ultra II, about $54,000 with 128 MB ram.
Where did this price come from??? We just bought an Ultra 2 with 64MB
of RAM for about $23,000 or less ( Sun never sells at given prices ) and
no we don't buy that many Sun's. We didn't opt for the second CPU which is
another $3000 so. The Ultra 2 was slightly more than a Sparc 20.
I don't think we took the Creator 3D graphics board either..