Linux as server OS? No?
Boris
www.tpc.org but couldn't find any...lol.
Boris
Eric Lee Green
>
> Where are TPC-C and TPC-D benchmark results for Linux based databases? I looked through
> www.tpc.org but couldn't find any...lol.
And your point is???
Anybody who specs based on some artificial benchmark, rather than by
setting up a test installation and feeding it their own particular
real-world workload, is an idiot. This is especially true for Linux,
where the necessary software to do a test is free or easily available
under 30 day demo license.
Nonwithstanding that, I wouldn't use Linux to run a multi-terabyte data
warehouse for a Fortune 500 company. That's why they invented Sun
Enterprise Servers (grin).
Wouldn't use NT for that either, BTW. It's a perfectly good file and
print server, decent workstation too, but I don't consider it stable
enough for mission-critical database work. Maybe okay for a small
business (like Linux), but not for big stuff. There's a reason why
NASDAQ uses Unix-based systems for their databases and not NT, despite
their very-high-profile "conversion" to NT (very high profile on the
microsoft.com site, that is). You see, when you set up a web site, you
can hide an unstable web server platform behind a cluster router (see
the microsoft.com site for various white papers on how to do this).
There's no such hiding a database machine failing.
--
Eric Lee Green http://members.tripod.com/e_l_green
mail: e_l_...@hotmail.com
^^^^^^^ Burdening Microsoft with SPAM!
Chris Costello
> Where are TPC-C and TPC-D benchmark results for Linux based databases? I looked through
> www.tpc.org but couldn't find any...lol.
You said that already, troll.
> Boris
--
|Chris Costello
|This is line 271 of .tags.
`--------------------------
W.A. Scheer
<0F0548DBE7EC846B.E80B88AB...@lp.airnews.net>...
[much snipiness]
>There's no such hiding a database machine failing.
Hmmm ... ever heard of DB2 clusters? Oracle Parallel Server? Works great on
NT ... and soon Linux!
Boris
Boris
Chris Costello <ch...@holly.dyndns.org> wrote in message
news:p_Fs3.123$DI3....@typhoon01.swbell.net...
Boris
W.A. Scheer <dans...@jps.net> wrote in message news:37b3...@news1.jps.net...
Boris
Steve Sheldon
>business (like Linux), but not for big stuff. There's a reason why
>NASDAQ uses Unix-based systems for their databases and not NT, despite
>their very-high-profile "conversion" to NT (very high profile on the
>microsoft.com site, that is). You see, when you set up a web site, you
The front end user-interface for the NASDAQ trading system runs on Windows
NT Workstation 4.0. There may also be a X-windows based client, but I might
be getting that confused with a different trading system.
I actually helped deploy this at a local financial company back in '97.
NASDAQ was very specific about what hardware we used, and how to install
the machines including driver versions, etc.
I'm curious to see you argue that trader's desktops are not mission
critical. That would be a good laugh after having spent 9 months on the
trading support desk.
--
Steve Sheldon email: she...@yuck.net
BSCS/MCSE url: http://www.sheldon.visi.com
Save a flag... Burn the Constitution.
Chris Costello
> Do you stink? Like Linux?
What, specifically, is that supposed to mean?
You are not a very well-written Perl script.
--
|Chris Costello
|Every bug you find is the last one.
`-----------------------------------
Chris Costello
x86 doesn't do good beyond 1 CPU as well as other
architectures do. What's your point?
> Boris
--
|Chris Costello
|All you need to know is the user interface. - J. Redford
`---------------------------------------------------------
Eric Lee Green
> As Mindcraft results showed Linux shit doesn't scale beyond one CPU.
You should really work on your language. Using words like "shit" shows a
distinct lack of professionalism and reflects adversely upon your cause.
As I am always lecturing the Linux advocates, if you wish to make a
point make it calmly, with the facts behind you, and with appropriate
and professional language. Otherwise you just hurt the cause you are
trying to help. (Hmm, that's an idea, are you really a Linux advocate
trying to make MS advocates look like total morons? Naw, nobody could be
that Byzantine :-).
Eric Lee Green
>
> Eric Lee Green <e_l_...@hotmail.com> writes:
>
> >business (like Linux), but not for big stuff. There's a reason why
> >NASDAQ uses Unix-based systems for their databases and not NT, despite
> >their very-high-profile "conversion" to NT (very high profile on the
> >microsoft.com site, that is). You see, when you set up a web site, you
>
> The front end user-interface for the NASDAQ trading system runs on Windows
> NT Workstation 4.0.
Exactly. That is the extent of NASDAQ's "conversion" to NT -- they
replaced a bunch of old Unix workstations with NT workstations. As I
said, NT makes a perfectly acceptable workstation or file and print
server. At the time that the conversion was spec'ed out, NT made sense,
much as it did in 1995 when I told my employer at the time that we had
better jump on the NT bandwagon or we were going to get creamed. Unix
was just too darned expensive, and the "free" Linux and FreeBSD were not
really ready for prime time.
But I note that you do not tell us that you helped with the conversion
of their database servers to NT. That's because you didn't -- their
database servers run on Unix. And will continue to do so for the near
future. Workstations can be swiftly rebooted without disturbing the
basic functioning of NASDAQ. The servers cannot.
Gary Hallock
> W.A. Scheer <dans...@jps.net> wrote in message news:37b3...@news1.jps.net...
> > Eric Lee Green wrote in message
> > <0F0548DBE7EC846B.E80B88AB...@lp.airnews.net>...
> > [much snipiness]
> > >There's no such hiding a database machine failing.
> >
> > Hmmm ... ever heard of DB2 clusters? Oracle Parallel Server? Works great on
> > NT ... and soon Linux!
> As Mindcraft results showed Linux shit doesn't scale beyond one CPU.
>
> Boris
And we all know what a total loser you are, Boris (or did you think I would
forget your comments about Linux users). You obviously feel so horrible about
your stinking life that you can only feel good by insulting others. Why don't
you get out and stay out!
Gary
je...@dementia.mishnet
>
>W.A. Scheer <dans...@jps.net> wrote in message news:37b3...@news1.jps.net...
>> Eric Lee Green wrote in message
>> <0F0548DBE7EC846B.E80B88AB...@lp.airnews.net>...
>> [much snipiness]
>> >There's no such hiding a database machine failing.
>>
>> Hmmm ... ever heard of DB2 clusters? Oracle Parallel Server? Works great on
>> NT ... and soon Linux!
>As Mindcraft results showed Linux shit doesn't scale beyond one CPU.
Actually, what it showed is that Linux doesn't scale well past
one NIC. For multiple-process concurrency, Linux can exploit
multiple processors quite nicely.
--
It helps the car, in terms of end user complexity and engineering,
that a car is not expected to suddenly become wood chipper at some |||
arbitrary point as it's rolling down the road. / | \
Seeking sane PPP Docs? Try http://penguin.lvcm.com
Steve Shaw
Eric Lee Green <e_l_...@hotmail.com> wrote
>
> Wouldn't use NT for that either, BTW. It's a perfectly good file and
> print server, decent workstation too, but I don't consider it stable
> enough for mission-critical database work. Maybe okay for a small
Yep, NT in a nutshell.
It's unfortunate that Microsoft feels compelled to puff NT as more than it
really is. Leaves a bad taste.
NT is an OK if unexceptional OS. Nice kernel, but too tightly-coupled to
disk for optimum performance. Given decent tuning, NT could be much better
than it is.
Steve S.
Boris
On www.tpc.org there are dozens of TPC-C benchmark results for NT (from 4 to 8 CPUs). But
not a single result for Linux. Correlating this info with results of Open Linux/NT
Benchmark conducted by Mindcraft, MS, and Linux engineers in zdnet testing lab (while
being arbitrated by zdnet engineers) leads to definite conclusion: Linux doesn't scale
beyond 1 CPU and cannot successfully compete with NT in server benchmarks.
Boris
Eric Lee Green <e_l_...@hotmail.com> wrote in message
news:6D7A15FA0E01EFAE.1111844B...@lp.airnews.net...
> Boris wrote:
> > As Mindcraft results showed Linux shit doesn't scale beyond one CPU.
>
> You should really work on your language. Using words like "shit" shows a
> distinct lack of professionalism and reflects adversely upon your cause.
> As I am always lecturing the Linux advocates, if you wish to make a
> point make it calmly, with the facts behind you, and with appropriate
> and professional language. Otherwise you just hurt the cause you are
> trying to help. (Hmm, that's an idea, are you really a Linux advocate
> trying to make MS advocates look like total morons? Naw, nobody could be
> that Byzantine :-).
>
Donovan Rebbechi
>Hmm..
>
>I remember reading an article in Computer World about NASDAQ converting
>thier DB servers over as well.
>
>They said that earlier last year they converted all the front ends and they
>were planning a major conversion of their
>DB servers.
Take a look at Kusnetzky's take on it.
They haven't even begun such a conversion yet. They will dabble their
toes in the water before they move everything over. Until they do some
toe-dabbling, it's early to speculate.
--
Donovan
je...@dementia.mishnet
>je...@dementia.mishnet () writes:
>
>>On Thu, 12 Aug 1999 15:25:15 -0700, Boris <boris...@pleasemovil.com> wrote:
>>>
>>>W.A. Scheer <dans...@jps.net> wrote in message news:37b3...@news1.jps.net...
>>>> Eric Lee Green wrote in message
>>>> <0F0548DBE7EC846B.E80B88AB...@lp.airnews.net>...
>>>> [much snipiness]
>>>> >There's no such hiding a database machine failing.
>>>>
>>>> Hmmm ... ever heard of DB2 clusters? Oracle Parallel Server? Works great on
>>>> NT ... and soon Linux!
>
>> one NIC. For multiple-process concurrency, Linux can exploit
>> multiple processors quite nicely.
>
Hardly. What we complained about was that the test was
a joke bought by MS. The subsequent and previous benchmarks
by both Mindcraft and others demonstrated that.
>
> That's all, nothing more, nothing less.
>
> Time to move on.
Then why are you dwelling on Mindcraft?
To not do so would only undermine your own Troll bait.
Gary Hallock
> OK. I'll put it politely.
> On www.tpc.org there are dozens of TPC-C benchmark results for NT (from 4 to 8 CPUs). But
> not a single result for Linux. Correlating this info with results of Open Linux/NT
> Benchmark conducted by Mindcraft, MS, and Linux engineers in zdnet testing lab (while
> being arbitrated by zdnet engineers) leads to definite conclusion: Linux doesn't scale
> beyond 1 CPU and cannot successfully compete with NT in server benchmarks.
>
> Boris
Hu??? Correlating what info? You just said there wasn't any. You are a total moron.
Just so other readers of this NG don't get the wrong idea. I do not go around insulting
people. But Boris has a reputation of using foul language to describe all Linux users.
He likes to come in here once in a while and make trouble. He seems to have no clue how
stupid he appears to other people. I think the term moron fits him.
Gary
Eric Lee Green
> On www.tpc.org there are dozens of TPC-C benchmark results for NT (from 4 to 8 CPUs). But
> not a single result for Linux. Correlating this info with results of Open Linux/NT
> Benchmark conducted by Mindcraft, MS, and Linux engineers in zdnet testing lab (while
> being arbitrated by zdnet engineers) leads to definite conclusion: Linux doesn't scale
> beyond 1 CPU and cannot successfully compete with NT in server benchmarks.
That's fair enough. NT cannot successfully compete with herring in
swimming benchmarks (grin). For the gory details see
http://members.tripod.com/e_l_green/benchmarketing101.html (heheh).
One thing to bear in mind: benchmarks only count insofar as they're
applicable to the problem set you're attacking. I agree that, at the
moment, Linux is a poor choice for a terabyte-sized data warehouse (I'd
suggest a Sun Enterprise Server there). On the other hand, that doesn't
make Linux useless as a small business server, and I'll point out that
NT doesn't scale well either (max of 8 processors at the moment, while
some of the "big iron" Unix boxes from Sun and SGI have as many as 64
processors... and even the latest'n'greatest PCI-64 I/O bus is bog-slow
compared with the I/O bandwidth available on that heavy-liftin'
hardware).
Eric Lee Green
> je...@dementia.mishnet () writes:
> > Actually, what it showed is that Linux doesn't scale well past
> > one NIC. For multiple-process concurrency, Linux can exploit
> > multiple processors quite nicely.
>
> Oh come on. What is showed was that the Linux advocates were wrong.
>
Not at all. I suggest that you read my response to Mindcraft (published
back when I was technical director at Linux Hardware Solutions), where I
basically stated that I was certain that, on this Microsoft-selected
benchmark, that NT would blow Linux away, but that Mindcraft's shoddy
conduct made it difficult to use the resulting benchmark numbers in
order to further improve Linux.
The PC Labs work verified two things: a) it verified that Mindcraft did
shoddy work, and b) it verified that NT blew Linux away on this
particular set of hardware and particular set of benchmarks. It also
gave Linux kernel developers plentiful data in order to locate the exact
problem areas in the Linux kernel, the most prominant of which was that
the network stack had a big lock and thus would only feed one network
card at a time (limiting bandwidth to that of one network card no matter
how many CPU's you had).
None of which has an iota to do with typical usage of Linux in the real
world. As I've pointed out, NT flunks benchmarks too (see
http://members.tripod.com/e_l_green/benchmarketing101.html for an
example of a benchmark that NT flunks). The only real benchmark is this:
Is it the most cost-effective solution for a particular task? If so, use
it. If not, don't use it. Simple, eh?
Eric Lee Green
> Ohwell, it's really unimportant. We're all aware that NT is used quite
> heavily in enterprise solutions. It's Linux we're still trying to find
> someone using for anything important...
Jay Jacobs running their stores with it isn't anything important?
The City of Garden Grove is minced cheesecake?
Half of the school districts in Louisiana and several in Mississippi
running their student information systems on Linux (and that's the most
mission critical system of all, because that detirmines state funding),
and that's not important?
I think we just discovered who's a troll.
je...@dementia.mishnet
>je...@dementia.mishnet () writes:
>
>>On Fri, 13 Aug 1999 02:17:29 GMT, Steve Sheldon <she...@visi.com> wrote:
>
>> a joke bought by MS. The subsequent and previous benchmarks
>> by both Mindcraft and others demonstrated that.
>
>Linux is faster than NT" followed by this awful sound of a whole bunch of
>people going "Doh!".
>
> The subsequent benchmarks by pcweek/mindcraft proved mindcraft's position.
>I'm not sure how you can spin this any other way.
No they didn't. They rather undermined them. What everyone
tends to ignore after that 'bought blowout benchmark' is
that the results in general weren't anything new. Just the
4:1 bit was.
[deletia]
je...@dementia.mishnet
>je...@dementia.mishnet () writes:
>
>> Hardly. What we complained about was that the test was
>> a joke bought by MS. The subsequent and previous benchmarks
>> by both Mindcraft and others demonstrated that.
>
>
>http://www.mindcraft.com/whitepapers/openbench1.html
I wasn't wrong. Even the last Mindcraft numbers support my position.
Bogus is still Bogus, less absurd numbers in the same general
direction don't validate bogus numbers.
je...@dementia.mishnet
>Heh.. typical. Linux advocates can't come up with real numbers, so they must
>mock
>all scientific benchmarks conducted by linux "experts" themselves.
Calling what Mindcraft does science is an insult to
all scientists and philosophers.
>
>How amusing, no?
>
>Chad
>
>Eric Lee Green <e_l_...@hotmail.com> wrote in message
>news:D0802DA8F8419C3E.257A256C...@lp.airnews.net...
>> Steve Sheldon wrote:
>> > je...@dementia.mishnet () writes:
>> > > Actually, what it showed is that Linux doesn't scale well past
>> > > one NIC. For multiple-process concurrency, Linux can exploit
>> > > multiple processors quite nicely.
>> >
>> > Oh come on. What is showed was that the Linux advocates were wrong.
>> >
>> > That's all, nothing more, nothing less.
>>
>> Not at all. I suggest that you read my response to Mindcraft (published
>> back when I was technical director at Linux Hardware Solutions), where I
>> basically stated that I was certain that, on this Microsoft-selected
>> benchmark, that NT would blow Linux away, but that Mindcraft's shoddy
>> conduct made it difficult to use the resulting benchmark numbers in
>> order to further improve Linux.
>>
>> The PC Labs work verified two things: a) it verified that Mindcraft did
>> shoddy work, and b) it verified that NT blew Linux away on this
>> particular set of hardware and particular set of benchmarks. It also
>> gave Linux kernel developers plentiful data in order to locate the exact
>> problem areas in the Linux kernel, the most prominant of which was that
>> the network stack had a big lock and thus would only feed one network
>> card at a time (limiting bandwidth to that of one network card no matter
>> how many CPU's you had).
>>
>> None of which has an iota to do with typical usage of Linux in the real
>> world. As I've pointed out, NT flunks benchmarks too (see
>> http://members.tripod.com/e_l_green/benchmarketing101.html for an
>> example of a benchmark that NT flunks). The only real benchmark is this:
>> Is it the most cost-effective solution for a particular task? If so, use
>> it. If not, don't use it. Simple, eh?
>>
>> --
>> Eric Lee Green http://members.tripod.com/e_l_green
>> mail: e_l_...@hotmail.com
>> ^^^^^^^ Burdening Microsoft with SPAM!
>
>
Chad
I remember reading an article in Computer World about NASDAQ converting
thier DB servers over as well.
They said that earlier last year they converted all the front ends and they
were planning a major conversion of their
DB servers.
Last I heard they were running a hybrid, they were keeping their unix system
around until the end of the year and they would
completely switch over after the millennium. They chose Dell servers
though... <shaking head> <sigh>.
Oh hey, the Dell servers are pretty cool though, they have an ATI Rage Pro
3D card in case you want to run Quake2 while
you're serving up your web pages!
Here's some articles, perhaps I was mistaken:
(I went to computerworld.com and did a search for NASDAQ between 2/01/1998
and 8/17/1999
Nasdaq taps NT for trade monitoring system
http://www.computerworld.com/home/news.nsf/all/9808171nasdaq
Nasdaq gets bullish on NT
http://www.computerworld.com/home/print.nsf/all/980928nasdaq
Nasdaq Targets Growth with Win NT Systems
http://www.computerworld.com/home/print.nsf/all/990517A692
NT Picked for Stock Analysis App
http://www.computerworld.com/home/print.nsf/all/98082463E2
The NT Factor (not really related to Nasdaq, but interesting nonetheless)
http://www.computerworld.com/home/print.nsf/all/9812148362
I've found Computer World to be pretty middle-of-the-road. I see as many
anti-MS articles as pro-MS articles and the same for linux and other OSs
Also, by the way, as far as multiterabyte NT systems, check out
http://terraserver.microsoft.com
Last I checked, 1.2TB of data and like 2.5TB of images. All running on 2
(last I checked) Compaq ProLiant 7000s with
SQL Server 7.0.
Actually, now that I look, they have Digital AlphaServer 8400's (again,
demonstrating NT's scalability and multi-platform capabilities)
They have seven cabinets with 46 each (one with 48) 9GB SCSI drives!!
(that's 3.330TB for you linux people out there)
Chad
Eric Lee Green <e_l_...@hotmail.com> wrote in message news:3C4799A2CC6D723
D.165A5736E6C0D28...@lp.airnews.net...
Steve Sheldon
>Steve Sheldon wrote:
>>
>> Eric Lee Green <e_l_...@hotmail.com> writes:
>>
>> >business (like Linux), but not for big stuff. There's a reason why
>> >NASDAQ uses Unix-based systems for their databases and not NT, despite
>> >their very-high-profile "conversion" to NT (very high profile on the
>> >microsoft.com site, that is). You see, when you set up a web site, you
>>
>> The front end user-interface for the NASDAQ trading system runs on Windows
>> NT Workstation 4.0.
>Exactly. That is the extent of NASDAQ's "conversion" to NT -- they
>replaced a bunch of old Unix workstations with NT workstations. As I
>said, NT makes a perfectly acceptable workstation or file and print
>server. At the time that the conversion was spec'ed out, NT made sense,
>much as it did in 1995 when I told my employer at the time that we had
>better jump on the NT bandwagon or we were going to get creamed. Unix
>was just too darned expensive, and the "free" Linux and FreeBSD were not
>really ready for prime time.
And still aren't, but that's a different issue.
>But I note that you do not tell us that you helped with the conversion
>of their database servers to NT. That's because you didn't -- their
>database servers run on Unix. And will continue to do so for the near
>future. Workstations can be swiftly rebooted without disturbing the
>basic functioning of NASDAQ. The servers cannot.
Well the reason I wasn't involved with the back end conversion was because
my client was a stock trading firm... not NASDAQ itself. We were shipped
hardware and software and instructions on how to install.
It's interesting that you ignored my comment about your not believing
trading systems are mission critical, but you make a coment about rebooting.
You don't seem to understand... If the computer does not respond to their
keystroke within 2 seconds they are blowing their top. Reboots are not an
acceptable solution.
I remained at the client for a while after the NASDAQ system was put in
place. We did not have any issues with the client piece running under NT,
however the middleware processes did fail occasionally. It was either
linked to something wrong on the Sun box or the link back to NASDAQ.
However unlike you I'm not going to try to claim that this meant the Unix
piece of the equation sucked.
Steve Sheldon
>On Thu, 12 Aug 1999 15:25:15 -0700, Boris <boris...@pleasemovil.com> wrote:
>>
>>W.A. Scheer <dans...@jps.net> wrote in message news:37b3...@news1.jps.net...
>>> Eric Lee Green wrote in message
>>> <0F0548DBE7EC846B.E80B88AB...@lp.airnews.net>...
>>> [much snipiness]
>>> >There's no such hiding a database machine failing.
>>>
>>> Hmmm ... ever heard of DB2 clusters? Oracle Parallel Server? Works great on
>>> NT ... and soon Linux!
>>As Mindcraft results showed Linux shit doesn't scale beyond one CPU.
> Actually, what it showed is that Linux doesn't scale well past
> one NIC. For multiple-process concurrency, Linux can exploit
> multiple processors quite nicely.
Oh come on. What is showed was that the Linux advocates were wrong.
That's all, nothing more, nothing less.
Time to move on.
Steve Sheldon
>Hmm..
>I remember reading an article in Computer World about NASDAQ converting
>thier DB servers over as well.
>They said that earlier last year they converted all the front ends and they
>were planning a major conversion of their
>DB servers.
>Last I heard they were running a hybrid, they were keeping their unix system
>around until the end of the year and they would
>completely switch over after the millennium. They chose Dell servers
>though... <shaking head> <sigh>.
This article below says they're using Unisys machines.
>Nasdaq Targets Growth with Win NT Systems
>http://www.computerworld.com/home/print.nsf/all/990517A692
Interesting.
I was only involved when the client frontends were put in in late '97.
This does seem to indicate a major convesion of legacy Unix to newer
Windows NT... Does this mean Eric Green is wrong?
I notice that nasdaq.com is an IIS 4 server, which is also interesting as
that server is a primary site for financial information.
Ohwell, it's really unimportant. We're all aware that NT is used quite
heavily in enterprise solutions. It's Linux we're still trying to find
someone using for anything important...
Steve Sheldon
>On Fri, 13 Aug 1999 02:17:29 GMT, Steve Sheldon <she...@visi.com> wrote:
>> Oh come on. What is showed was that the Linux advocates were wrong.
> Hardly. What we complained about was that the test was
> a joke bought by MS. The subsequent and previous benchmarks
> by both Mindcraft and others demonstrated that.
Well actually I was referring to the incessant chattering... "Of course
Linux is faster than NT" followed by this awful sound of a whole bunch of
people going "Doh!".
The subsequent benchmarks by pcweek/mindcraft proved mindcraft's position.
I'm not sure how you can spin this any other way.
>>
>> That's all, nothing more, nothing less.
>>
>> Time to move on.
> Then why are you dwelling on Mindcraft?
You don't seem to undrestand. Those of us in
comp.os.ms-windows.nt.advocacy are acting as a support group for Linux
zealots. We just want to help. Admitting you were wrong is the first step
to admitting you have a problem.
Steve Sheldon
> Hardly. What we complained about was that the test was
> a joke bought by MS. The subsequent and previous benchmarks
> by both Mindcraft and others demonstrated that.
Oh, in case you missed your dose of reality check...
http://www.mindcraft.com/whitepapers/openbench1.html
Now just admit you were wrong, and we'll move on...
Chad
mock
all scientific benchmarks conducted by linux "experts" themselves.
How amusing, no?
Chad
Eric Lee Green <e_l_...@hotmail.com> wrote in message
news:D0802DA8F8419C3E.257A256C...@lp.airnews.net...
> Steve Sheldon wrote:
> > je...@dementia.mishnet () writes:
> > > Actually, what it showed is that Linux doesn't scale well past
> > > one NIC. For multiple-process concurrency, Linux can exploit
> > > multiple processors quite nicely.
> >
> > Oh come on. What is showed was that the Linux advocates were wrong.
> >
> > That's all, nothing more, nothing less.
>
> Not at all. I suggest that you read my response to Mindcraft (published
> back when I was technical director at Linux Hardware Solutions), where I
> basically stated that I was certain that, on this Microsoft-selected
> benchmark, that NT would blow Linux away, but that Mindcraft's shoddy
> conduct made it difficult to use the resulting benchmark numbers in
> order to further improve Linux.
>
> The PC Labs work verified two things: a) it verified that Mindcraft did
> shoddy work, and b) it verified that NT blew Linux away on this
> particular set of hardware and particular set of benchmarks. It also
> gave Linux kernel developers plentiful data in order to locate the exact
> problem areas in the Linux kernel, the most prominant of which was that
> the network stack had a big lock and thus would only feed one network
> card at a time (limiting bandwidth to that of one network card no matter
> how many CPU's you had).
>
> None of which has an iota to do with typical usage of Linux in the real
> world. As I've pointed out, NT flunks benchmarks too (see
> http://members.tripod.com/e_l_green/benchmarketing101.html for an
> example of a benchmark that NT flunks). The only real benchmark is this:
> Is it the most cost-effective solution for a particular task? If so, use
> it. If not, don't use it. Simple, eh?
>
W.A. Scheer
keep the linvocates from misusing benchmark data ... now Boris has to chime
in.
For the record, people ... please learn something about the art/science of
benchmarking before waxing eloquent about what various benchmark results
"mean". The results of TPC benchmarks (or any other benchmark) do not
"prove" that a particular OS is superior than another one. One reason why
there are no Linux TPC benchmarks AS OF YET is because a TPC benchmark is a
long, expensive process requiring several hundred thousand dollars in
equipment. Many companies simply choose not to participate for that reason
alone. Note the lack of any TPC/D results posted on SCO Unixware or
Openserver. Does this mean that "they suck"? Hardly. How 'bout the lack of
ANY benchmarks on Gateway Servers? Same thing.
+++++++++++++++++++++++++++++++++++++++++++++++++++++
Boris wrote in message <37b38009$0$84...@nntp1.ba.best.com>...
>OK. I'll put it politely.
A.T.Z.
> Where are TPC-C and TPC-D benchmark results for Linux based databases? I looked through
> www.tpc.org but couldn't find any...lol.
>
> Boris
Great troll.
Chad
news:slrn7r7eh...@dementia.mishnet...
must
> >mock
> >all scientific benchmarks conducted by linux "experts" themselves.
Oh here we go with more confuciousism (sp? wow that's a toughy!) or Zen
computing...
"Science is not science, is the documented study of philosophy" right?
So computers are not silicon and copper/alluminum, they really only exist in
our heart
and because linux is free, it's easier on our heart and we should all be
using linux
on our imaginary computers in our hearts eh?
<sigh>
The fact is, the numbers show it.. Linux doesn't scale beyond one CPU. The
Redhat
"engineers" admitted that the linux kernel 2.2.x had major multithreading
problems in the
networking stack (I don't believe it was just TCP/IP, I think it was the
whale networking module)
so you could put 1000 processors in there and it wouldn't matter.
Yet linux is ready for the enterprise *choke* yeah, sure.
Chad
Eric Lee Green
> Heh.. typical. Linux advocates can't come up with real numbers, so they must
> mock
> all scientific benchmarks conducted by linux "experts" themselves.
Typical. NT advocates can't win if it's turned into a question of "what
is the most cost-effective platform that performs a particular task"
(such as, say, departmental firewall), so they turn pointy-haired boss
on us and throw out benchmarks that are applicable only to some other
task.
Repeat after me: Those who repeatedly choose a more expensive solution
when a less expensive solution will do are doomed to eventual
unemployment. If your commute needs are served by a Toyota, it is
irrational to buy a Mercedes. Yeah, it's good ego-boo to say "Yeah, I
have a Mercedes", but in a business environment it's all about dollars
and cents, not ego-boo.
Right now, Linux is a Toyota, and NT is an SUV. NT will do some things
that Linux won't do, but Linux is cheaper and sips less gas (runs on
smaller computers). If you need four network cards and four processors,
use NT. If you need a firewall, use Linux. Most cost-effective solution
for a particular task, no?
When you're fired because you continue to insist on using a
more-expensive solution to a particular problem when a less-expensive
solution exists, don't bother asking me for a job (even though we're
hiring). I'm not interested in inflexible poltroons who are more
interested in egoboo than in getting the work done in the most
cost-effective manner.
Grant Fischer
>je...@dementia.mishnet () writes:
>
>> Hardly. What we complained about was that the test was
>> a joke bought by MS. The subsequent and previous benchmarks
>> by both Mindcraft and others demonstrated that.
>
> Oh, in case you missed your dose of reality check...
>
>http://www.mindcraft.com/whitepapers/openbench1.html
>
I always found it funny that one of the key requirements was
that the machine have 4 NIC's. They were willing to experiment
on software, # CPU's, etc, but never the variable that was
bringing out the specific problem in the Linux kernel.
You've got to remember that their original benchmark, which
caused so much fuss, had the Linux performance collapsing
diastrously. They also made claims about trying to seek
help, and not succeeding. They initially denied running
the tests at an MS lab, and the guy who posted for help
seemed to be an MS or ex-MS employee. Various Linux
people had suggestions on minimum changes that should
be implemented before benchmarking -- they wanted those
simple changes ruled out as the cause of the bad performance.
In other words, at least give it a fair chance.
They've not been able to justify that original benchmark,
despite how much they'd like us to forget it.
Yes, NT beat Linux in benchmark #2 & #3. The two tests were close
enough together to show that there wasn't much the Linux
guys could do without fixing the network stack lock.
Mindcraft was vindicated in not having any Linux guys on-site
for test #2 -- there wasn't much they could have done.
It took the c't benchmark to bring out the reality check,
showing the specific cause.
--
Grant Fischer (gfischer at the domain hub.org)
Eric Lee Green
> The fact is, the numbers show it.. Linux doesn't scale beyond one CPU.
Depends on the task.
> Yet linux is ready for the enterprise *choke* yeah, sure.
As a departmental file and print server, yes. As a small business or
small government database server, yes (city of Garden Grove CA, half the
school districts in Louisiana, Enhanced Software Technologies Inc.). As
a web server, EMAIL server, and internet server, yes (see
http://www.deja.com and about 90% of the Internet Service Providers out
there). As a corporate database server, no. Use a "real" Unix there.
Real Unix's still have features, like hot-swap PCI, hot-swap CPU cards,
logical volume management, scaling beyond 8 CPU's, REAL logging
filesystems (that don't need hours to do a filesystem check on
terabyte-sized volumes if you have an unclean shutdown!) that "toy"
computers running NT and Linux can currently only dream about.
I don't know why you continue to insist that based on benchmarks Linux
is unsuited to tasks where it is currently performing quite adequately
in real life. You're like the guy who insists that bumblebees cannot fly
because it violates all known laws of aerodynamics. Yet they do.
Ian Smith
>
> je...@dementia.mishnet () writes:
>
> > Hardly. What we complained about was that the test was
> > a joke bought by MS. The subsequent and previous benchmarks
> > by both Mindcraft and others demonstrated that.
>
> Oh, in case you missed your dose of reality check...
>
> http://www.mindcraft.com/whitepapers/openbench1.html
>
See ya, wouldn't want to be ya!
--
--------------------------------------------------------
Ian Smith Linux Help: alt.os.linux
uk.comp.os.linux
--------------------------------------------------------
Ian Smith
>
> As Mindcraft results showed Linux shit doesn't scale beyond one CPU.
Boris, why not step outside and play go f*** yourself?
Joseph T. Adams
: there). As a corporate database server, no. Use a "real" Unix there.
: Real Unix's still have features, like hot-swap PCI, hot-swap CPU cards,
: logical volume management, scaling beyond 8 CPU's, REAL logging
: filesystems (that don't need hours to do a filesystem check on
: terabyte-sized volumes if you have an unclean shutdown!) that "toy"
: computers running NT and Linux can currently only dream about.
You probably know this already, but for those who don't: Linux does
support 64-bit files on 64-bit hardware; ext3 will probably support
journaling; the journaling (but very fast) XFS filesystem created by
SGI should be available for Linux within a year or so; SMP and the
TCP/IP stack are under heavy development, partly to remove the
bottlenecks exposed during the Mindcraft tests; and, finally,
enterprise-class databases don't generally use the filesystem, and
have their own schemes for clustering so that there doesn't need to be
any single point of failure.
For the very largest terabyte-size databases I'd probably choose Sun
hardware, but for almost anything else I would not hesitate to
recommend clustered Linux systems if a more cost-effective solution
did not present itself.
: I don't know why you continue to insist that based on benchmarks Linux
: is unsuited to tasks where it is currently performing quite adequately
: in real life. You're like the guy who insists that bumblebees cannot fly
: because it violates all known laws of aerodynamics. Yet they do.
The great thing is that Linux is performing well in tasks that no one
thought it would ever be able to do two or three years ago.
I believe that Linux, and the open-source model in general, will prove
to be more scalable than any closed-source systems. This has not yet
been proven, but I think it will be.
Joe
John Hughes
Eric Lee Green <e_l_...@hotmail.com> wrote in message
news:CEDE04B5C1CFFF73.683114A2...@lp.airnews.net...
> Eric Lee Green http://members.tripod.com/e_l_green
> mail: e_l_...@hotmail.com
> ^^^^^^^ Burdening Microsoft with SPAM!
By applying a mixture of OS for different tasks your creating integration
problems not to mention needing staff trained in multiple OS'. Because Linux
is free doesnt mean its going to be less expensive. If you think this then
your being pretty shallow and I personally wouldnt want to work for someone
that that anyway!
John Hughes
Grant Fischer <gfische...@hub.org> wrote in message
news:slrn7r87ha.ci3....@dogbert.cibg.tdbank.ca...
> On Fri, 13 Aug 1999 04:15:54 GMT, Steve Sheldon <she...@visi.com> wrote:
> >je...@dementia.mishnet () writes:
> >
> >> Hardly. What we complained about was that the test was
> >> a joke bought by MS. The subsequent and previous benchmarks
> >> by both Mindcraft and others demonstrated that.
> >
> > Oh, in case you missed your dose of reality check...
> >
> >http://www.mindcraft.com/whitepapers/openbench1.html
> >
>
Cmon the C't benchmark was just as unfair as the Mindcraft #1.
CGI on IIS isnt an option as you have more scalable options.
To this date I havent saw a benchmark which shows Linux scaling with todays
technologies.
John Hughes
> Chad wrote:
> > The fact is, the numbers show it.. Linux doesn't scale beyond one CPU.
>
> Depends on the task.
>
> > Yet linux is ready for the enterprise *choke* yeah, sure.
>
> As a departmental file and print server, yes. As a small business or
> small government database server, yes (city of Garden Grove CA, half the
> school districts in Louisiana, Enhanced Software Technologies Inc.). As
> a web server, EMAIL server, and internet server, yes (see
> http://www.deja.com and about 90% of the Internet Service Providers out
> there).
Bollocks. Get a list of ISP's and do the netcraft test on them. Its all FUD.
We did the test in the UK and a very small amount were using Linux.
As a corporate database server, no. Use a "real" Unix there.
> Real Unix's still have features, like hot-swap PCI, hot-swap CPU cards,
> logical volume management, scaling beyond 8 CPU's, REAL logging
> filesystems (that don't need hours to do a filesystem check on
> terabyte-sized volumes if you have an unclean shutdown!) that "toy"
> computers running NT and Linux can currently only dream about.
>
> I don't know why you continue to insist that based on benchmarks Linux
> is unsuited to tasks where it is currently performing quite adequately
> in real life. You're like the guy who insists that bumblebees cannot fly
> because it violates all known laws of aerodynamics. Yet they do.
>
Boris
Boris
Gary Hallock <gha...@ibm.net> wrote in message news:37B39445...@ibm.net...
> Boris wrote:
>
> > OK. I'll put it politely.
> > On www.tpc.org there are dozens of TPC-C benchmark results for NT (from 4 to 8 CPUs).
But
> > not a single result for Linux. Correlating this info with results of Open Linux/NT
> > Benchmark conducted by Mindcraft, MS, and Linux engineers in zdnet testing lab (while
> > being arbitrated by zdnet engineers) leads to definite conclusion: Linux doesn't scale
> > beyond 1 CPU and cannot successfully compete with NT in server benchmarks.
> >
> > Boris
>
> Hu??? Correlating what info? You just said there wasn't any. You are a total
moron.
>
> Just so other readers of this NG don't get the wrong idea. I do not go around
insulting
> people. But Boris has a reputation of using foul language to describe all Linux users.
> He likes to come in here once in a while and make trouble. He seems to have no clue
how
> stupid he appears to other people. I think the term moron fits him.
>
> Gary
>
Chris Costello
> You are cretin.
Based on what?
--
|Chris Costello
|Command, n.:
| Statement presented by a human and accepted by a computer in
|such a manner as to make the human feel as if he is in control.
`--------------------------------------------------------------------
Donn Miller
> By applying a mixture of OS for different tasks your creating integration
> problems not to mention needing staff trained in multiple OS'. Because Linux
> is free doesnt mean its going to be less expensive. If you think this then
> your being pretty shallow and I personally wouldnt want to work for someone
> that that anyway!
Not necessarily. You can, for example, run an X server on an NT
machine to run apps on a remote UNIX server running X. The fact
that practically any platform out there can run X makes for
seamless integration between those two platforms running X.
Also note that with NT, you can run something like Cygwin-32, and
you can have both platforms more compatible in terms of coding,
although your NT platforms would be UNIX-biased. NT is supposed
to be compatible with UNIX systems up to a certain extent,
because certain versions of NT have a POSIX subset included.
(They have a name for this, what is it, Integrix or something?)
-- Donn
Boris
> > On www.tpc.org there are dozens of TPC-C benchmark results for NT (from 4 to 8 CPUs).
But
> > not a single result for Linux. Correlating this info with results of Open Linux/NT
> > Benchmark conducted by Mindcraft, MS, and Linux engineers in zdnet testing lab (while
> > being arbitrated by zdnet engineers) leads to definite conclusion: Linux doesn't scale
> > beyond 1 CPU and cannot successfully compete with NT in server benchmarks.
>
> swimming benchmarks (grin). For the gory details see
> http://members.tripod.com/e_l_green/benchmarketing101.html (heheh).
>
> One thing to bear in mind: benchmarks only count insofar as they're
> applicable to the problem set you're attacking. I agree that, at the
> moment, Linux is a poor choice for a terabyte-sized data warehouse (I'd
> suggest a Sun Enterprise Server there).
NT (with its middleware: MTS, MSMQ) and SQL Server 7 make excellent database/application
server.
I remember reading that SAP said that MS SQL 7 power would satisfy 90% of SAP's customers.
Latest TPC/C benchmarks on NT are very impressive: 40,000 trans. per minute with
price-performance ratio < 20.
http://www.tpc.org/results/individual_results/Compaq/compaq.8500.99071202.es.pdf
For comparison, here are price performance numbers for other systems with ~ 40,000
throughput:
1. Compaq Alpha system: ~50 ( actually, TPC throughput is lower: around 30,000 - for this
system).
http://www.tpc.org/results/individual_results/Compaq/compaq.gs60e.99081101.es.pdf
2. AS400: ~ 70
http://www.tpc.org/results/individual_results/IBM/ibm.as400.99060701.es.pdf
In this range (medium) NT has 2.5 - 3.5 times price/performance advantage.
>On the other hand, that doesn't
> make Linux useless as a small business server, and I'll point out that
> NT doesn't scale well either (max of 8 processors at the moment, while
> some of the "big iron" Unix boxes from Sun and SGI have as many as 64
> processors... and even the latest'n'greatest PCI-64 I/O bus is bog-slow
> compared with the I/O bandwidth available on that heavy-liftin'
> hardware).
NT does scale do 8 CPUs on Intel platform (see results above).
High-end PC servers have multiple PCI buses.
Also, NT will be making in-roads into mission-critical market VERY soon: Nasdaq pilot
project will use NT on Unisys servers.
http://www.infoworld.com/cgi-bin/displayStat.pl?/pageone/news/features/ntworld/98ntworld.b
riefs.htm
Boris
Boris
> keep the linvocates from misusing benchmark data ... now Boris has to chime
> in.
>
> For the record, people ... please learn something about the art/science of
> benchmarking before waxing eloquent about what various benchmark results
> "mean". The results of TPC benchmarks (or any other benchmark) do not
> "prove" that a particular OS is superior than another one. One reason why
> there are no Linux TPC benchmarks AS OF YET is because a TPC benchmark is a
> long, expensive process requiring several hundred thousand dollars in
> equipment.
I understand that Linux developers have money now. For example, IBM accepted Linux as
server platform. What's several hundred thousand $ in equipment for IBM? Nothing. However,
IBM publishes NT TPC results but NOT any Linux results. Why is that? Because this thing
(Linux) doesn't scale.
Many companies simply choose not to participate for that reason
> alone. Note the lack of any TPC/D results posted on SCO Unixware or
> Openserver. Does this mean that "they suck"? Hardly. How 'bout the lack of
> ANY benchmarks on Gateway Servers? Same thing.
Gateway servers? Not mainstream servers. High-volume NT server vendors: Compaq, IBM, HP,
Dell - do publish TPC numbers for NT on regular basis (www.tpc.org ). SCO Unixware? Very
low end.
Boris
Boris
~20,000 for 4-way server and ~40,000 for 8-way server.
http://www.tpc.org/results/individual_results/Compaq/compaq.pl5500.99071301.es.pdf
http://www.tpc.org/results/individual_results/Compaq/compaq.8000.99071201.es.pdf
Linux is completely different story...lol.
Boris
Chris Costello <ch...@holly.dyndns.org> wrote in message
news:QwIs3.210$DI3....@typhoon01.swbell.net...
> In comp.os.linux.advocacy Boris <boris...@pleasemovil.com> wrote:
> > As Mindcraft results showed Linux shit doesn't scale beyond one CPU.
>
> x86 doesn't do good beyond 1 CPU as well as other
> architectures do. What's your point?
>
> > Boris
>
>
>
> --
> |Chris Costello
> |All you need to know is the user interface. - J. Redford
> `---------------------------------------------------------
Boris
Boris
Chris Costello <ch...@holly.dyndns.org> wrote in message
news:kAZs3.47$bZ1....@typhoon01.swbell.net...
> In comp.os.linux.advocacy Boris <boris...@pleasemovil.com> wrote:
> > You are cretin.
>
> Based on what?
>
> > Boris
> > Gary Hallock <gha...@ibm.net> wrote in message news:37B39445...@ibm.net...
> >> Boris wrote:
> >>
> >> > OK. I'll put it politely.
> >> > On www.tpc.org there are dozens of TPC-C benchmark results for NT (from 4 to 8
CPUs).
> > But
> >> > not a single result for Linux. Correlating this info with results of Open Linux/NT
> >> > Benchmark conducted by Mindcraft, MS, and Linux engineers in zdnet testing lab
(while
> >> > being arbitrated by zdnet engineers) leads to definite conclusion: Linux doesn't
scale
> >> > beyond 1 CPU and cannot successfully compete with NT in server benchmarks.
> >> >
Chris Costello
> Based on my analysis of contents of his brain: it's made of shit.
> Boris
What is a brain surgeon doing on USENET?
--
|Chris Costello
|Gotta run, the cat's caught in the printer.
`-------------------------------------------
Chris Costello
> ~20,000 for 4-way server and ~40,000 for 8-way server.
> http://www.tpc.org/results/individual_results/Compaq/compaq.pl5500.99071301.es.pdf
> http://www.tpc.org/results/individual_results/Compaq/compaq.8000.99071201.es.pdf
> Linux is completely different story...lol.
I wasn't talking about NT, I meant x86 in general. If
Microsoft can produce a version of NT that works as good on
another architecture, such as Sparc, as x86, scaling beyond one
CPU would be better than what would be seen under x86.
--
|Chris Costello
|All that glitters has a high refractive index.
`-----------------------------------------------
Stephen Edwards
: In comp.os.linux.advocacy Boris <boris...@pleasemovil.com> wrote:
: > Based on my analysis of contents of his brain: it's made of shit.
: > Boris
: What is a brain surgeon doing on USENET?
Ask Hobbyist... *grin*
--
.-----.
|[_] :| Stephen S. Edwards II | http://www.primenet.com/~rakmount
| = :| "Well, next time I'm in dire need of an opinion of absolutely
| | no value, you'll be the first to know." -- Chad Mulligan
|_..._|
je...@dementia.mishnet
'cgi' isn't the only option on Linux either.
>
>To this date I havent saw a benchmark which shows Linux scaling with todays
>technologies.
Ditto for NT. That's one of the 'interesting' bits about Mindcraft.
It doesn't bear any resemblance to sites that serve that kind of
volume.
--
It helps the car, in terms of end user complexity and engineering,
that a car is not expected to suddenly become wood chipper at some |||
arbitrary point as it's rolling down the road. / | \
Seeking sane PPP Docs? Try http://penguin.lvcm.com
Steve Sheldon
>On Fri, 13 Aug 1999 04:15:54 GMT, Steve Sheldon <she...@visi.com> wrote:
>>je...@dementia.mishnet () writes:
>>
>>> Hardly. What we complained about was that the test was
>>> a joke bought by MS. The subsequent and previous benchmarks
>>> by both Mindcraft and others demonstrated that.
>>
>> Oh, in case you missed your dose of reality check...
>>
>>http://www.mindcraft.com/whitepapers/openbench1.html
> I wasn't wrong. Even the last Mindcraft numbers support my position.
> Bogus is still Bogus, less absurd numbers in the same general
> direction don't validate bogus numbers.
Oh man, you are more of a spin master than most politicians.
Boris
It has very powerful features to help achieve that goal.
1. High performance and good scalability on modern hardware (latest result on www.tpc.org
show that NT scales to 8 CPUs on TPC C benchmark).
2. Decent uptime.
NT stability increased greatly since NT4 was first released. NT servers stay up for months
without re-boot.
3. NTFS with journaling support, advanced security, and very large disk/file support.
4. Middleware with transaction support: MTS, MSMQ.
For example, let's consider customer buying a computer over the phone. Customer calls in
and orders a PC. Sales representative creates an order and submits it to database server.
Let's say database server or network are down at the moment. MSMQ will ensure completing
the transaction when problem is fixed. As far as sales representative is concerned the
order was submitted for processing. But because MSMQ doesn't require that database server
is reachable at the time of the transaction, record will be queued on sales workstation
and sent to server when connection becomes available. Because this mechanism is
transaction based queued record will not be deleted until confirmation received from the
server. In Win2K (COM+) this mechanism will be transparent for application developer who
will just write standard COM component; OS will take care of the rest.
5. Very advanced and relatively extremely cheap MS SQL Server 7.
Just compare prices of Oracle (or Sybase) DBMS with MS SQL 7 in TPC/C results sheets. The
first cost 300 - 400K while MS SLQ Server 7 costs ~30K.
NT Server 4 has some problems/limitations. In my opinion those are the following:
1. Lack of 64 bit version.
Due in 2000 (if I recall correctly).
2. No good clustering support.
3. No process quotas.
I've seen many times that "pig" application eats up all memory or creates too many objects
(open handles) and brings OS to virtual stop. Win2K addresses this issue with new
mechanism: job object. Job-wide or process-wide quotas in a job include memory, cpu time,
security and GUI limits.
4. No structured process tree.
Actually, by using trivial device driver your application can maintain structured process
tree. For example, if you use watchdog service to monitor health of some apps and this
thing determines that one of monitored apps died, it's possible to kill all child
processes before restarting the main app.
Win2K solves that problem via job object as well: it's possible to terminate all processes
in a job.
Boris
Steve Shaw <sbsh...@email.msn.com> wrote in message news:uN2VYhS5#GA.409@cpmsnbbsa02...
>
> Eric Lee Green <e_l_...@hotmail.com> wrote
> >
> > Wouldn't use NT for that either, BTW. It's a perfectly good file and
> > print server, decent workstation too, but I don't consider it stable
> > enough for mission-critical database work. Maybe okay for a small
> > business (like Linux), but not for big stuff.
>
> Yep, NT in a nutshell.
>
> It's unfortunate that Microsoft feels compelled to puff NT as more than it
> really is. Leaves a bad taste.
>
> NT is an OK if unexceptional OS. Nice kernel, but too tightly-coupled to
> disk for optimum performance. Given decent tuning, NT could be much better
> than it is.
>
> Steve S.
>
Steve Sheldon
>Typical. NT advocates can't win if it's turned into a question of "what
>is the most cost-effective platform that performs a particular task"
>(such as, say, departmental firewall), so they turn pointy-haired boss
>on us and throw out benchmarks that are applicable only to some other
>task.
http://www.microsoft.com/ntserver/nts/exec/compares/ntlinux.asp
How exactly do you determine cost-effectiveness?
>Repeat after me: Those who repeatedly choose a more expensive solution
>when a less expensive solution will do are doomed to eventual
>unemployment. If your commute needs are served by a Toyota, it is
>irrational to buy a Mercedes. Yeah, it's good ego-boo to say "Yeah, I
>have a Mercedes", but in a business environment it's all about dollars
>and cents, not ego-boo.
It's not quite as simple as that.
>Right now, Linux is a Toyota, and NT is an SUV. NT will do some things
>that Linux won't do, but Linux is cheaper and sips less gas (runs on
>smaller computers). If you need four network cards and four processors,
>use NT. If you need a firewall, use Linux. Most cost-effective solution
>for a particular task, no?
Not sure I'd use Linux as a firewall. I value my data too much for that.
But if I was setting up a little server for the Agronomy department at
college, sure...
>When you're fired because you continue to insist on using a
>more-expensive solution to a particular problem when a less-expensive
>solution exists, don't bother asking me for a job (even though we're
>hiring). I'm not interested in inflexible poltroons who are more
>interested in egoboo than in getting the work done in the most
>cost-effective manner.
That's all nice, but it is more directed at the Linux advocates than the NT
advocates.
We're just interested in solutions that work... you're the ones with the
Religion to spread.
Steve Sheldon
>>http://www.mindcraft.com/whitepapers/openbench1.html
>I always found it funny that one of the key requirements was
>that the machine have 4 NIC's. They were willing to experiment
>on software, # CPU's, etc, but never the variable that was
>bringing out the specific problem in the Linux kernel.
It's not an uncommon configuration. Also the goal was to test the
throughput of the infrastructure and as such you need suitable
infrastructure to push.
>You've got to remember that their original benchmark, which
>caused so much fuss, had the Linux performance collapsing
>diastrously. They also made claims about trying to seek
And this was attributed to a bug in the kernel code as I recall.
>help, and not succeeding. They initially denied running
Which were shown to be true.
>the tests at an MS lab, and the guy who posted for help
There was no such denial.
>seemed to be an MS or ex-MS employee. Various Linux
I thought you said they didn't ask for help?
>people had suggestions on minimum changes that should
>be implemented before benchmarking -- they wanted those
>simple changes ruled out as the cause of the bad performance.
>In other words, at least give it a fair chance.
They just did.
>They've not been able to justify that original benchmark,
>despite how much they'd like us to forget it.
They just did.
>Yes, NT beat Linux in benchmark #2 & #3. The two tests were close
>enough together to show that there wasn't much the Linux
>guys could do without fixing the network stack lock.
>Mindcraft was vindicated in not having any Linux guys on-site
>for test #2 -- there wasn't much they could have done.
Hmm.
>It took the c't benchmark to bring out the reality check,
>showing the specific cause.
Odd, the c't benchmarks I saw seemed to be everything you are accusing the
Mindcraft ones of being.
Or are you speaking of a different set of benchmarks? Do you have a link?
Gary Hallock
> You are cretin.
>
Fine, resort to name calling it you want. You still haven't answered my question. What info
are you correlating? You are drawing a conclusion for ZERO data. You said that yourself.
Gary
Gary Hallock
> Based on my analysis of contents of his brain: it's made of shit.
>
Yep, same old Boris! You have yet to figure out that I do know a thing or two about
computers. You are the one that introduced yourself as believing that, and I quote:
> There's that Linux mob which causes strong desire to vomit. Basically it's a crowd of
> shit-headed engineers who think they are capable to change software industry with their
> ridiculous freeware.
> They plain lie about performance, stability, and other issues of their fetish (Linux) as
> compared to other OSes (NT for example). Typical Linux engineer is a loser who cannot
> compete on widely accepted terms but claims that he has smthg better: Linux. When actual
> bencmark results show that Linux is a pile of crap those people shout about fowl play,
> etc. Psychologically it's quite understandable: losers are afraid to publicly accept the
> fact that they are losers.
And this personal attack was totally unprovked. I responded to this post at the time, to
demonstrate that I am not a loser. But of course, you cut and ran rather than respond.
Boris, it doesn't matter what you say. You have shown yourself to be a total idiot and and
with no credibility. Whay do you do this to yourself?
Gary
je...@dementia.mishnet
>je...@dementia.mishnet () writes:
>
>>On Fri, 13 Aug 1999 04:15:54 GMT, Steve Sheldon <she...@visi.com> wrote:
>>>je...@dementia.mishnet () writes:
>>>
>>>> Hardly. What we complained about was that the test was
>>>> a joke bought by MS. The subsequent and previous benchmarks
>>>> by both Mindcraft and others demonstrated that.
>>>
>>> Oh, in case you missed your dose of reality check...
>>>
>>>http://www.mindcraft.com/whitepapers/openbench1.html
>
>> I wasn't wrong. Even the last Mindcraft numbers support my position.
>
>> Bogus is still Bogus, less absurd numbers in the same general
>> direction don't validate bogus numbers.
>
> Oh man, you are more of a spin master than most politicians.
No, you're the one indulging in all the spin. If you
can tell us what the results of Mindcraft's toy
benchmark actually buy the real webmaster, you might
have a point.
Donovan Rebbechi
>http://www.microsoft.com/ntserver/nts/exec/compares/ntlinux.asp
>
>How exactly do you determine cost-effectiveness?
Let's take a look at the misleading claims in the report:
(1) The studies are described as "independent"
(2) NT performs "up to 650% better" as a webserver. These
results are fudged by comparing Apache/CGI against IIS/ASP. IOW, it's
hardly an unbiased comparison ( PHP or mod_perl would be a better
choice on linux if performance was an issue )
(3) More "statistics" are obtained with an equally slanted CGI vs
ISAPI comparison
(4) Static Webpage performance comparison is misleading, because it
assumes that both servers have unlimited available bandwidth.
The fact is that either server can saturate a dedicated T3.
(5) Their "price/performance" ratio assumes that the machines have
unlimited bandwidth. The truth is that msot webservers wil not
have the kind of bandwidth to utilise NT's apparent performance
advantage.
In the real world, you have X amount of bandwidth, and need the
lowest cost solution to utilise that bandwidth. You don't have
unlimited bandwidth.
Notice that they don't factor bandwidth costs into the comparison.
The reason for this is that the annual cost of a couple of
dedicated T3 connections dwarfs the cost of their servers.
( in fact two dedicated T3 lines wil cost more *in the first
month* than either server ) Factoring in bandwidth costs will
kill NT's "advantage" (esp since it's expressed as a percentage )
(6) The evaluation assumes that the NT box only has 60 client access
licenses, but in practise, the clients are not *all* actively
transferring files concurrently ( as in the test ). You will
need several more clients ( and hence client access licenses )
to utilise the bandwidth in the same manner as the test.
(7) In the "It's about more than performance", we see some humour:
Reliability:
99.9% uptime ? is this a bug or a feature ?
note that this a gaurantee against *unplanned* downtime,
( it works out at 9hrs a year) which is why MS don't
give the URLs.
Anyone familiar with linux is also familiar with the fact
that 9 hours unscheduled downtime/year isn't something to
boast about.
Security:
Linux apparently "inherits the security flaws of UNIX",
for example it's easy to gain root access via "poorly
written applications" ( like IIS ? ) The fact is that
code written by amateurs doesn't run as root on a linux
box.
"No resolution or path for fixing bugs with
clear accountability":
THere is no "clear accountability" with closed software.
Read the EULA. In practice, security bugs in OSS are
rapidly fixed.
"No centralised security..."
Hilarious, considering that Win 200 is using kerberos
which has been available on linux for years.
"More prone to security bugs"
No evidence provided. The reference says that UNIX and NT are
both secure, but NT has ACLs which are cool, and NT is "easy
to admin". The former I agree with, the latter is contentious.
NT is only reliable and secure if the admin is competent.
TCO:
Claims are unsubstantiated. The study that showed NT to have
lower "TCO" than UNIX made several contrived assumptions.
"Ease of Use": A monkey can't competently admin an NT box.
Sorry.
--
Donovan
Eric Lee Green
> Eric Lee Green <e_l_...@hotmail.com> wrote in message news:32A7F1DA36F8C239.948E87A2...@lp.airnews.net...
> > As a departmental file and print server, yes. As a small business or
> > small government database server, yes (city of Garden Grove CA, half the
> > school districts in Louisiana, Enhanced Software Technologies Inc.). As
> > a web server, EMAIL server, and internet server, yes (see
> > http://www.deja.com and about 90% of the Internet Service Providers out
> > there).
>
> Bollocks. Get a list of ISP's and do the netcraft test on them. Its all FUD.
>
> We did the test in the UK and a very small amount were using Linux.
Sorry, I was talking about in major markets, not in some third world
island with delusions of grandeur where they think boiling food is the
best way to cook it and where there are few ISP's because of a state
monopoly phone system that insists upon charging per-minute fees for
Internet access. We long ago figured out here in the 'States that state
monopolies for phone service were a bad idea, that's why we have over 40
million people connected to the Internet and why your damp foggy island
has, what, 8 people connected to the Internet? (heheh).
Brief survey of the ISP's in Shreveport, Louisiana, a typical mid-sized
city in the United States, finds a total of 18 ISP's, of which 12 are
using Linux. The others are using BSDI or Solaris, by the way, so don't
get your hopes up about NT. In fact, not a single major ISP in the
United States is using Windows NT. AOL? No. AT&T? No. BellSouth? No.
U.S. West? No. IBM? No. Mindspring? No. If you can name one (1) major
ISP (as in, available in at least two states if not nationwide) that
uses Windows NT, I'll happily mail you a Cajun Cooking cookbook and
assortment of spices so that you can cook some real food (grin). (Note
-- no boiling involved, though, so you may have to figure out how to
saute' and carmelize, but I'm sure you can do it!).
je...@dementia.mishnet
>gfische...@hub.org (Grant Fischer) writes:
>
>>>http://www.mindcraft.com/whitepapers/openbench1.html
>
>>I always found it funny that one of the key requirements was
>>that the machine have 4 NIC's. They were willing to experiment
>>on software, # CPU's, etc, but never the variable that was
>>bringing out the specific problem in the Linux kernel.
>
> It's not an uncommon configuration. Also the goal was to test the
>throughput of the infrastructure and as such you need suitable
>infrastructure to push.
Except they didn't test that infastructure with anything
resembling what a site of that size would be using.
Even the machines on MS.com don't sport 4 NICs.
[deletia]
Eric Lee Green
> > more-expensive solution to a particular problem when a less-expensive
> > solution exists, don't bother asking me for a job (even though we're
> > hiring). I'm not interested in inflexible poltroons who are more
> > interested in egoboo than in getting the work done in the most
> > cost-effective manner.
>
> problems
Integration problems? With a Linux firewall? The firewall is transparent
to client OS's, it doesn't even KNOW what client OS is out there!
> not to mention needing staff trained in multiple OS'.
Like they're so hard to find. We went to a Linux user's group meeting
and hired some kid off the floor who seemed like he knew what he was
talking about, and hey, you know what? He does just fine! I did the
master IT plan, but he does just fine at implementing it, and guess what
-- he is a HECK of a lot cheaper than a MSCE. The fact is that Linux is
hot amongst the kids, and the kids work for CHEAP. A few old hands like
myself (no, I don't have a bushy white beard and suspenders :-) are all
that's needed to put them to work doing Great Things.
>Because Linux
> is free doesnt mean its going to be less expensive.
That is absolutely correct. For example, if you have an NT consultant in
your town and you don't have a Linux consultant, it doesn't make sense
to call in a Linux consultant from far away and pay travel fees etc. So
even though, on an absolute cost basis, NT is more expensive than Linux,
the recurring costs would make Linux the more expensive solution IN THAT
SITUATION.
On the other hand, the only way to figure that out is to do a
cost-benefit analysis. Dismissing a solution out-of-hand because it is
not the One True Operating System is ridiculous. Or do you seriously
insist that Windows NT is The One True Operating System? If so, I pity
you, because eventually that inflexibility will catch up with you and
you'll be in the same position as the old Cobol-heads on February 1,
2000 -- out of a job, walking the streets, and complaining wistfully
about how these new-fangled thingies aren't as good as MVS/CICS and
COBOL on good ole' IBM 'frames.
In my case, I did a cost-benefit analysis (and yes, personnel
considerations WERE a factor, we would have needed to hire NT-trained
personnel to supplement our Unix-trained personnel if we'd gone NT), and
found that Linux was the most cost-effective solution FOR OUR PARTICULAR
SITUATION. Gosh, is that such a hard notion to capture?
Idiots who insist that their solution is the "best" solution without
doing a cost-benefit analysis deserve the bad rep that they get.
Unfortunately, they do the rest of us a disservice too, by giving the
industry as a whole a bad rep when they push ill-suited solutions onto
customers who must then contend with buggy unreliable service for high
prices.
Eric Lee Green
> Eric Lee Green <e_l_...@hotmail.com> writes:
> >Typical. NT advocates can't win if it's turned into a question of "what
> >is the most cost-effective platform that performs a particular task"
> >(such as, say, departmental firewall), so they turn pointy-haired boss
> >on us and throw out benchmarks that are applicable only to some other
> >task.
>
> http://www.microsoft.com/ntserver/nts/exec/compares/ntlinux.asp
Ah yes! Note the careful applications of the principals of FUD (see
http://members.tripod.com/e_l_green/fud101.html ). They are especially
keen on benchmarketing (see where I use benchmarks to "prove" that
herring are better than Windows
http://members.tripod.com/e_l_green/benchmarketing101.html ). They do
especially large helpings of FUD Method #2 (Inventing Weaknesses), using
Distraction Method #1 (Sandwiching). I have been meaning to update the
FUD101 Howto to bring it up-to-date with Microsoft's latest developments
in its FUD attacks against Linux (especially Benchmarketing, an
insidious case of FUD#1 combined with FUD#2 using Distraction Technique
#1 and #5 heavily), and this page is a treasure-trove of examples to put
into the FUD101 Howto. Hmm, I released the FUD101 Howto back in November
of 1998, detailing the tactics that Microsoft was going to take against
the Linux community in the upcoming year. It's interesting how relevant
it still is today, a year later. The only thing I really missed was
Benchmarketing, I don't know why, I've been around this industry long
enough to know about Benchmarketing.
> How exactly do you determine cost-effectiveness?
Simple: You gather together all the solutions that may solve your
problem. You test them with real-world workloads (as vs. some artificial
benchmarks) to see how they respond under load. You then take the
remaining contenders and add up the costs involved. Those costs might
be: installation costs (how much will it cost to either hire or contract
someone to install this solution?), up-front costs (license fees and
such), recurring costs (how much will it cost to hire or contract people
to maintain this solution? How many people will be needed? How much
hassle will backing up the systems be? How difficult will it be to keep
the system up and running, will there be a difference in the number of
times we need to call in consultants to fix things?), and future upgrade
costs (how much will it cost me to upgrade to the next version of this
product when it reaches its lifespan and the vendor no longer supports
it?). You then choose the solution that a) meets the real-world needs,
b) performs adequately under load, c) is least expensive, from a TCO
point of view, including both up-front costs, recurring costs, and
future upgrade costs.
This is standard operating procedure in most industries -- choosing the
correct tool for the job. It is pathetic that this industry has so many
unqualified people who, by dint of passing a few idiotic tests, think
they're God's gift to Information Technology. Hate to tell you, guy, but
there's more to being an IT specialist than knowing how to tweak obscure
registry settings using "regedit". Have you ever even TAKEN an
Operations Research course?!
> >Right now, Linux is a Toyota, and NT is an SUV. NT will do some things
> >that Linux won't do, but Linux is cheaper and sips less gas (runs on
> >smaller computers). If you need four network cards and four processors,
> >use NT. If you need a firewall, use Linux. Most cost-effective solution
> >for a particular task, no?
>
> Not sure I'd use Linux as a firewall. I value my data too much for that.
See what I mean? Rigidity. Dismissing a possible solution without
testing it first to see if it will do the job. I'd fire you in a minute.
> We're just interested in solutions that work... you're the ones with the
> Religion to spread.
Oh poop. I'm the guy who recommended to my boss that we move to Windows
NT, remember? Because, at the time, it was a choice of expensive
proprietary Unix boxes and Windows NT, and Windows NT won in terms of
usability and cost. When Linux came along I dismissed it out-of-hand as
a "hobby platform" until Red Hat Software released their commercial
Linux. It took six months of testing and two pilot sites before I
finally admitted that yeah, Linux was as stable as the proprietary Unix
we'd been using, and was cheaper than NT. While it did not do everything
that NT did (and still doesn't), it met our needs, for a lower cost.
Which allowed us to under-bid our competition while still making more
profit. Which was the whole point. Money. Not religion.
Things have changed, but some old Cobol-heads still insist that Windows
NT and Visual BASIC are the solution to all problems. (Whoops, that
should have been "MVS/CICS and COBOL", me bad!).
Eric Lee Green
> TCO:
>
> Claims are unsubstantiated. The study that showed NT to have
> lower "TCO" than UNIX made several contrived assumptions.
And in fact, the latest data from IDC shows that TCO for Unix in a 1,000
user environment is close to $1,000 per user per year cheaper than for
Windows NT. Source: 1998 IDC Report on TCO. Unix cost: $5,394 per user.
NT cost: $6,180 per user. This covers all software costs, software
maintenance contracts, salaries of all IT people involved in maintaining
the solution, and costs of all external consultants needed to keep the
solution running.
In other words, Microsoft lied.
But then, why am I not surprised?
These costs are the primary reason why Egghead.com moved from Windows NT
to Sun Enterprise Servers. Said Tom Collins, chief technology officer of
Egghead, "Maintenance was the main issue. The costs and effort to
maintain the boxes were too much. We were continually rebooting, which
always took servers out of the loop." They are using E3500 and E4500
Solaris boxes now, and report that "they just work."
Gary Hallock
>
> IBM's Big Blue(tm), carries somewhere around 5200 processors.
> *SPOOGE*.
>
> Okay, I'm done. I want a Cray TE6.. :)
>
> ~Geoff W.
Yep. And guess what we used to design, simulate, and test Big Blue. -
AIX. No NT in site.
Gary
Eric Lee Green
> Anyway, we are not talking about ISP's here. We are talking about your
> desire to take over the corporate IT world with Linux by replacing Windows
> NT and Novell Netware. Let's stick to the topic, shall we?
Excellent, dear sir! Excellent! Attributing to me ideas that I have
never expressed, while easily dismissing an application where Linux has
clearly shown itself to be useful (hmm, http://www.deja.com is another
example). My hat off to you.
As I'll repeat: The only time to use Linux is when it's the most
cost-effective solution to a particular task. The same applies to
Windows NT, Novell, and whatever else you want to consider. I have even
STATED, in real WORDS, that there are applications where NT is the
preferred solution -- such as when you have a network to saturate with
static content from four network cards (grin).
Which I have repeated over and over ad nauseum for the past forty
messages that I have addressed to you.
Hmm, I'm going to have to add another FUD Distraction Method, "the straw
man". Which is where an artificial argument attributed to the
competition is set up (the artificial argument being that "Linux should
replace NT in all applications"), and then demolished.
Of course, I'm not made out of straw, and I'm not a typical idiot like
that Matthew person that you keep making a fool out of, so try to be a
little more inventive in the future.
je...@dementia.mishnet
>je...@dementia.mishnet () writes:
>
>>On Fri, 13 Aug 1999 22:11:14 GMT, Steve Sheldon <she...@visi.com> wrote:
>>>je...@dementia.mishnet () writes:
>>>
>>>> Bogus is still Bogus, less absurd numbers in the same general
>>>> direction don't validate bogus numbers.
>>>
>>> Oh man, you are more of a spin master than most politicians.
>
>> No, you're the one indulging in all the spin. If you
>> can tell us what the results of Mindcraft's toy
>> benchmark actually buy the real webmaster, you might
>> have a point.
>
>
> We *HAVE* servers configured very similarly to the ones Mindcraft tested
>here in our data center. Except we use Compaq, not Dell, and only two NICs
>not four, except we use gigabit over fibre not 100baseT over cat-5.
>
> Why do you keep claiming otherwise?
>
> This is what Mindcraft says about the issue:
>
>---
>Why did you use a four-processor server?
That's not the point. The usage pattern is. Also, it
wasn't the four processors (as C't demonstrated) that
were the problem but the 4 NIC's.
Hobbyist ©
sagacious words :
> Chris Costello <ch...@holly.dyndns.org> wrote:
>
> : In comp.os.linux.advocacy Boris <boris...@pleasemovil.com> wrote:
>
> : > Based on my analysis of contents of his brain: it's made of shit.
>
> : > Boris
>
> : What is a brain surgeon doing on USENET?
>
> Ask Hobbyist... *grin*
OK, OK, I know that my type is uncommon around these parts. A few
of us do have more than average interest in computing.
--
Ali ©
Eric Lee Green
> >> Not sure I'd use Linux as a firewall. I value my data too much for that.
>
> >See what I mean? Rigidity. Dismissing a possible solution without
> >testing it first to see if it will do the job. I'd fire you in a minute.
>
>
> I'm dismissing the possible solution because I've already seen it
> implemented numerous time and fail. In each and every case the company
> scrapped Linux and went with a professional firewall solution.
>
I see. So estinc.com's firewall is a professional firewall solution. So
shreve.net's firewall is a professional firewall solution. It's only
coincidence that when they start up, they give the LILO: boot prompt and
then say "Linux 2.x.x" ....
As for those "professional firewall solutions", do you know how many of
them are actually based upon FreeBSD/OpenBSD/NetBSD? Hint: it's the ones
that aren't based on Linux or QNX :-). As far as I know only Cisco has
tried to create a stand-alone firewall that is not based upon one of the
well-known embedded Internet operating systems, and I seem to recall
quite a few security exploits of their PIX firewall.
Here's some more of those "professional firewall solutions"...
http://www.borderware.com/borderware.html -- gosh, they say "a
hardened version of Unix" (some *BSD, they don't say which)...
http://www.watchguard.com FireBox II -- hmm, says "Linux 2.0
kernel"...
http://www.tlogic.com/ -- "embedded BSD" (again doesn't say which
one)
http://www.netscreen.com -- appears to be based on embedded QNX (they
don't say exactly, but the specs are similar). I like this box. If I
needed a firewall for more than a T1, this is the one I'd want. I
especially like the hardware-accelerated IPSec, which would come in
handy for telecommuting.
Anyhow: I agree that once you get up above a T1, you probably want
something with more horsepower than the typical Linux firewall. On the
other hand, an old Pentium 166 is currently handling our T1 firewall
duties just fine under Linux, and was cheap (well, "free", since we
replaced all of our Pentium 166 workstations several months ago, and the
old machines are cluttering our warehouse because discarding things is
such a paperwork PITA). You're saying that our firewall isn't really
working? Hmm. The only time it's ever crashed is when a co-worker kicked
the network card out of its slot, and I don't think anybody has logged
into it for the past three months. It just works. Whoops, I logged into
it the other day to set up an IPSec VPN between our LAN at work and my
LAN at home. Rebooted it too, to update to the 2.0.36 kernel with the
latest FreeS/WAN software (IPSec for Linux). Sorry about that, I almost
fibbed there :-).
Eric Lee Green
>
> Eric Lee Green <e_l_...@hotmail.com> wrote in message
> news:CEDE04B5C1CFFF73.683114A2...@lp.airnews.net...
> Again, spewing forth the narrowmindedness and shortsightedness...
>
> cost-effective implies TOC. TOC of linux in most configurations is much
> higher
> seeing as how it has to be nursed along and requires a full-time admin just
> to
> keep the linux box running.
Really? I hadn't noticed that I was a full-time admin :-). While I was
originally hired as a sysadmin, once I finished deploying Linux there
was no need for a full-time system administrator. Thus I am now
primarily a software engineer, working on elegant solutions to Unix tape
backup problems. I spend maybe fifteen minutes a day reading system logs
and such, and the rest of the time doing my real job (software
engineering). The network "just works". The only user intervention is
stuffing a tape in the server every night, then reading the log in the
morning to make sure it actually backed up. (It's never failed to back
up, but I'm paranoid about my data). Note that all user data is backed
up at that same time, since all user data resides on the server and not
on workstations (thanks to the magic of NIS, NFS, and DHCP, each and
every workstation is byte-for-byte identical, with the exception of
whatever minor changes are needed to adjust for the video card). In
short, it's as close to a zero administration installation as I've ever
encountered.
And guess what -- Linux is a Unix derived operating system, and shares
the same TCO advantage over Windows NT as Unix does. Read the 1998 IDC
report on TCO. Read it and weap. $5,394 vs. $6,180 (Unix vs. NT).
What's this BS you keep spouting about TCP?
> > Repeat after me: Those who repeatedly choose a more expensive solution
> > when a less expensive solution will do are doomed to eventual
> > unemployment. If your commute needs are served by a Toyota, it is
> > irrational to buy a Mercedes. Yeah, it's good ego-boo to say "Yeah, I
> > have a Mercedes", but in a business environment it's all about dollars
> > and cents, not ego-boo.
>
> More like Yugo and Mercedes. Yugo may cost less, but lasts half as long and
> requires
> frequent repairs and more money over time.
Really? are you going into that old "Linux is Communist" thing again?
Sheesh. (Comparing Linux to a Communist-built car... next thing you
know, you'll be comparing Linus Torvalds to Hitler!).
> need a firewall use linux? hah! I wouldn't trust a one-node network to
> protection
> from a linux firewall (or firecheesecloth would be more analogous)
Really? When I was running the networks for Linux Hardware Solutions we
certainly got a lot of attacks upon our networks from the Windows
"script kiddies", but nobody ever breached it. EST Inc. doesn't get as
much traffic, but our firewall there has never been breached either.
There's no secret, guy -- all it requires is removing all software that
can be used to breach the system, and then keep up with security
updates.
If Linux is such swiss cheese as a firewall, somebody can certainly get
into 4.9.114.169 and tell us all what line 3 of the file "testalarm.py"
in my home directory says. Heheh. Of course, all that lamer Microsoft
script kiddies seem capable of doing is tossing all sorts of IP
fragments and stuff at servers, which I guess could crash me if I were
running an older version of Linux. (I won't tell you even what OS I'm
running, grin -- am I running Linux 2.0.37? Linux 2.2.11? FreeBSD 3.2?
FreeBSD 4.0? Telnetting to my box won't tell you either, though port 80
may have a clue, grin). Oh whoops, so sorry, that doesn't give you a
clue either. Oh darn.
> The "best" firewalls are either hardware based or software based on NT
> (Raptor, FireWall-1, etc)
> or some variation of BSD.
WatchGuard is swiss cheese? (They're Linux-based).
I'd have to disagree with your assessment of NT-based firewalls. The
best firewalls I have seen there for big pipes (i.e. bigger than T1)
have hardware IPSec accelleration (for VPN's). My favorite appears to
run a version of QNX as its operating system, actually, though a couple
of others seem to be BSD-based. BTW, have you figured out what my system
at home is running?
> How 'bout when you're fired because you had to explain to the CEO
> that your site went down or got hacked because you wanted to
> save a few bucks up front ?
FUD FUD FUD. All the script kiddies in the world could not get in to
crack.linuxppc.org, all they could do was ping-flood it, and you're
saying that it got hacked? Man. Corner a Microsoft advocate, and all
they do is lie on you. Sheesh.
Do you have a SINGLE VERIFIABLE EXAMPLE of a properly-configured Linux
firewall EVER being cracked? Or are you just pulling bull**** out of
your a**? (Nevermind, I know the answer to that question). Oh, I don't
count "Oh, I set up a Linux firewall and it got cracked" to be
verifiable -- you've already proven you're capable of telling some
whoppers.
je...@dementia.mishnet
>gfis...@amblehub.amble.net (Grant Fischer) writes:
>
>>The 4-NIC benchmark might be relevent to some large file
>>servers. I can't see a typical small office using such a machine.
>
> Uhh... what's a small office have to do with this?
>
> The purpose was to show what happened on data center sized hardware... i.e.
>a file server for a company with 3,000 employees.
Which brings us back to the greater likelihood of multiple
CPU's or even multiple machines rather than multiple NICs.
[deletia]
je...@dementia.mishnet
>je...@dementia.mishnet () writes:
>
>>On Sat, 14 Aug 1999 01:46:28 GMT, Steve Sheldon <she...@visi.com> wrote:
>>> You still just don't seem to get it.
>>>
>>> We *HAVE* servers configured very similarly to the ones Mindcraft tested
>
>>>here in our data center. Except we use Compaq, not Dell, and only two NICs
>>>not four, except we use gigabit over fibre not 100baseT over cat-5.
>>>
>>> Why do you keep claiming otherwise?
>>>
>>> This is what Mindcraft says about the issue:
>>>
>>>---
>>>Why did you use a four-processor server?
>
>> That's not the point. The usage pattern is. Also, it
>> wasn't the four processors (as C't demonstrated) that
>> were the problem but the 4 NIC's.
>
>
>
> You're a slipperly little critter. At every turn you have a completely new
>excuse.
No, this is just weak attempt to attack the person rather than the arugment.
je...@dementia.mishnet
>The fact of the matter is, NT was able to scale to four CPUs, Linux wasn't.
>
>We can argue all day about wether or not someone _really_ needs 4 NICs.
No, it didnt' scale to 4 NICs and you are just trying to
intentionally confuse the issue.
[deletia]
je...@dementia.mishnet
>Eric Lee Green <e_l_...@hotmail.com> writes:
>
[deletia]
>>preferred solution -- such as when you have a network to saturate with
>>static content from four network cards (grin).
>
>I've been trying to figure out if a file server handling 2,000 user
>connections is small or large.
It's not that simple actually. There are some file server
applications that only serve 200 people but would likely
require more effective thruput than some solutions that
'handle 2000' people.
Geoff Winans
<e_l_...@hotmail.com> wrote:
>One thing to bear in mind: benchmarks only count insofar as they're
>applicable to the problem set you're attacking. I agree that, at the
>moment, Linux is a poor choice for a terabyte-sized data warehouse (I'd
>suggest a Sun Enterprise Server there). On the other hand, that doesn't
>make Linux useless as a small business server, and I'll point out that
>NT doesn't scale well either (max of 8 processors at the moment, while
>some of the "big iron" Unix boxes from Sun and SGI have as many as 64
>processors... and even the latest'n'greatest PCI-64 I/O bus is bog-slow
>compared with the I/O bandwidth available on that heavy-liftin'
>hardware).
>
>--
>Eric Lee Green http://members.tripod.com/e_l_green
> mail: e_l_...@hotmail.com
> ^^^^^^^ Burdening Microsoft with SPAM!
Damn straight - There are SGI Boxen out there, with 128 processors, as
well. :) (I think Sun's biggest system, offers 128, maybe
more...correct me on that)...
Paul 'Z' EwandeŠ
Boris <boris...@pleasemovil.com> a écrit dans le message : 37b497e6$0
Boris, excuse me to ask this question, but why, since to be able to post
level headed material like below, you feel the need to come across as a jerk
by insulting people and using foul language ?
You accomplish very little by behaving like that. You do some much more
effective advocacy by posting technical arguments than insults. IMO
anyway...
Paul 'Z' Ewande
$2...@nntp1.ba.best.com...
Geoff Winans
>
>Also, NT will be making in-roads into mission-critical market VERY soon: Nasdaq pilot
>project will use NT on Unisys servers.
>http://www.infoworld.com/cgi-bin/displayStat.pl?/pageone/news/features/ntworld/98ntworld.b
>riefs.htm
>
>Boris
>
>
Dear god, save us all.
~Geoff
Geoff Winans
>
>Bollocks. Get a list of ISP's and do the netcraft test on them. Its all FUD.
>
>We did the test in the UK and a very small amount were using Linux.
I'm sure he ment in the way of UNIX in general, not nessicarily Linux.
Quit reading so deep into things. You make yourself look anal that
way.
~Geoff
Steve Sheldon
>On Fri, 13 Aug 1999 22:11:14 GMT, Steve Sheldon <she...@visi.com> wrote:
>>je...@dementia.mishnet () writes:
>>
>>> Bogus is still Bogus, less absurd numbers in the same general
>>> direction don't validate bogus numbers.
>>
>> Oh man, you are more of a spin master than most politicians.
> No, you're the one indulging in all the spin. If you
> can tell us what the results of Mindcraft's toy
> benchmark actually buy the real webmaster, you might
> have a point.
You still just don't seem to get it.
We *HAVE* servers configured very similarly to the ones Mindcraft tested
here in our data center. Except we use Compaq, not Dell, and only two NICs
not four, except we use gigabit over fibre not 100baseT over cat-5.
Why do you keep claiming otherwise?
This is what Mindcraft says about the issue:
---
Why did you use a four-processor server?
As stated in the answer to the previous question, we were originally hired
by Microsoft to show the performance of Linux and Windows NT server on an
enterprise-class system. While many in the Linux community may not have
experience using computers with four 400 MHz Xeon processors with 2 MB of
cache and 1 GB of RAM, such servers are common in large enterprises.
----
In fact nearly every single one of your little excuses has been addressed:
http://www.mindcraft.com/whitepapers/openbench1-faq.html
Sigh
Steve Sheldon
>On Fri, 13 Aug 1999 22:16:57 GMT, Steve Sheldon wrote:
>>http://www.microsoft.com/ntserver/nts/exec/compares/ntlinux.asp
>>
>>How exactly do you determine cost-effectiveness?
>Let's take a look at the misleading claims in the report:
>(1) The studies are described as "independent"
Irrelevant.
>(2) NT performs "up to 650% better" as a webserver. These
> results are fudged by comparing Apache/CGI against IIS/ASP. IOW, it's
> hardly an unbiased comparison ( PHP or mod_perl would be a better
> choice on linux if performance was an issue )
Do you have any links to benchmarks to support this?
>(3) More "statistics" are obtained with an equally slanted CGI vs
> ISAPI comparison
Sounds like one shouldn't use CGI.
>(4) Static Webpage performance comparison is misleading, because it
> assumes that both servers have unlimited available bandwidth.
> The fact is that either server can saturate a dedicated T3.
Gigabit ethernet is faster than T3, which is where most companies are
moving for backbones. Granted you wouldn't want to saturate your hole
backbone, but...
>(5) Their "price/performance" ratio assumes that the machines have
> unlimited bandwidth. The truth is that msot webservers wil not
> have the kind of bandwidth to utilise NT's apparent performance
> advantage.
Relevant if you need it.
> In the real world, you have X amount of bandwidth, and need the
> lowest cost solution to utilise that bandwidth. You don't have
> unlimited bandwidth.
Again relevant if you need the bandwidth.
> Notice that they don't factor bandwidth costs into the comparison.
> The reason for this is that the annual cost of a couple of
> dedicated T3 connections dwarfs the cost of their servers.
> ( in fact two dedicated T3 lines wil cost more *in the first
> month* than either server ) Factoring in bandwidth costs will
> kill NT's "advantage" (esp since it's expressed as a percentage )
Since the idea here is that your site has a certain level of traffic and in
either solution you will be purchasing the same level of internet bandwidth,
this is irrelevant.
>(6) The evaluation assumes that the NT box only has 60 client access
> licenses, but in practise, the clients are not *all* actively
> transferring files concurrently ( as in the test ). You will
> need several more clients ( and hence client access licenses )
> to utilise the bandwidth in the same manner as the test.
Hmm, you switched gears here... Now you went from internet to corporate
file serving. Yet nearly all of your comments above were not entirely
relevant to internal networks.
>(7) In the "It's about more than performance", we see some humour:
> Reliability:
>
> 99.9% uptime ? is this a bug or a feature ?
> note that this a gaurantee against *unplanned* downtime,
> ( it works out at 9hrs a year) which is why MS don't
> give the URLs.
> Anyone familiar with linux is also familiar with the fact
> that 9 hours unscheduled downtime/year isn't something to
> boast about.
Yes, Linux is used to more like 100 hours...
> Security:
> "No resolution or path for fixing bugs with
> clear accountability":
> THere is no "clear accountability" with closed software.
> Read the EULA. In practice, security bugs in OSS are
> rapidly fixed.
This is irrelevant.
In practice, security bugs in commercial software are rapidly fixed, and
the vendor has accountability for the problem.
> "No centralised security..."
> Hilarious, considering that Win 200 is using kerberos
> which has been available on linux for years.
Considering NIS doesn't even scale properly on linux, I'd be curious to
hear of any companies using kerberos on a large scale with linux.
The availability of something does not mean it works...
> "More prone to security bugs"
> No evidence provided. The reference says that UNIX and NT are
> both secure, but NT has ACLs which are cool, and NT is "easy
> to admin". The former I agree with, the latter is contentious.
> NT is only reliable and secure if the admin is competent.
Unix is only reliable and secure if the admin is competent. Again another
irrelevant point.
> TCO:
> Claims are unsubstantiated. The study that showed NT to have
> lower "TCO" than UNIX made several contrived assumptions.
> "Ease of Use": A monkey can't competently admin an NT box.
> Sorry.
While I agree. The largest cost in admin is generally the very basic
add/move/change stuff. NT is far easier to train someone in than Linux.
Steve Sheldon
>John Hughes wrote:
>> Bollocks. Get a list of ISP's and do the netcraft test on them. Its all FUD.
>>
>> We did the test in the UK and a very small amount were using Linux.
>Brief survey of the ISP's in Shreveport, Louisiana, a typical mid-sized
>city in the United States, finds a total of 18 ISP's, of which 12 are
>using Linux. The others are using BSDI or Solaris, by the way, so don't
>get your hopes up about NT. In fact, not a single major ISP in the
I really don't care about using NT for an ISP. ISP has needs completely
different than the corporate IT world which NT was targetted for.
As far as number of ISP's using Linux. If you do a weighted average on
that... i.e. weight the ISP by the number of customers you will find that
Linux is in fact not used by any of the major players.
An ISP with 500 customers is quite different from one with 15,000 would you
not agree? Shouldn't the ISP with 15,000 customers count as being more
important than the one with 500?
Anyway, we are not talking about ISP's here. We are talking about your
desire to take over the corporate IT world with Linux by replacing Windows
NT and Novell Netware. Let's stick to the topic, shall we?
Grant Fischer
>gfische...@hub.org (Grant Fischer) writes:
>
>>>http://www.mindcraft.com/whitepapers/openbench1.html
>
>>I always found it funny that one of the key requirements was
>>that the machine have 4 NIC's. They were willing to experiment
>>on software, # CPU's, etc, but never the variable that was
>>bringing out the specific problem in the Linux kernel.
>
> It's not an uncommon configuration. Also the goal was to test the
>throughput of the infrastructure and as such you need suitable
>infrastructure to push.
Does that make it a common configuration? What perentage of
servers do you run with 4 NIC's? We usually run our big Sun servers
with two for redundancy, and maybe a third for a dedicated
net-backup network. We've got two boxes with more; they are in an
HA cluster, so have two more NIC's apiece as dedicated heartbeat
links. I used to think in terms of multiple NIC's to get the
best out of ethernet and routers, but todays switching networks
have changed all that.
The 4-NIC benchmark might be relevent to some large file
servers. I can't see a typical small office using such a machine.
I look for benchmarks to show the strengths and weaknesses
of various solutions. The cynic in me sees that Mindcraft tried
to give the appearance of varying the configuration, but
not on the number of NIC's, which just so happened to
be the key to having NT perform better.
>
>>You've got to remember that their original benchmark, which
>>caused so much fuss, had the Linux performance collapsing
>>diastrously. They also made claims about trying to seek
>
> And this was attributed to a bug in the kernel code as I recall.
Yes, very good point. Sort of something that you might want
to see if there's a work-around, eh?
>
>>help, and not succeeding. They initially denied running
>
> Which were shown to be true.
I didn't mean that they tried to ask for help and couldn't get
the words out. I meant they claimed that they didn't
receive any offers of help, which is clearly wrong.
[Don't be obtuse. They claimed that they didn't get any help,
but didn't respond to the person offering advice on the
Internet, and tried to call the RedHat help line for
benchmark assistance. They should have approached RH
differently.]
>
>>the tests at an MS lab, and the guy who posted for help
>
> There was no such denial.
Maybe I'm mistaken. I seem to recall that in the early
days of the second test, the Linux guys pressured MC
to allow them onsite, and upon being denied there were
theories that the tests were being run at an MS lab.
It makes me think that any MS lab connection hadn't
been cleared up.
Sorry if I got that wrong.
>>seemed to be an MS or ex-MS employee. Various Linux
>
> I thought you said they didn't ask for help?
See above.
>
>>people had suggestions on minimum changes that should
>>be implemented before benchmarking -- they wanted those
>>simple changes ruled out as the cause of the bad performance.
>>In other words, at least give it a fair chance.
>
> They just did.
They screwed very basic things that could have had an
effect, including if I recall they turned on a wide-links
parameter in Samba that definitely slows it down.
Getting the basics right is essential.
>>They've not been able to justify that original benchmark,
>>despite how much they'd like us to forget it.
>
> They just did.
Where? You seem to be aware that the original disastrous
performance was due to a kernel bug. They should have at
least worked with RH to try to get that resolved, among
other things.
>
>>Yes, NT beat Linux in benchmark #2 & #3. The two tests were close
>>enough together to show that there wasn't much the Linux
>>guys could do without fixing the network stack lock.
>>Mindcraft was vindicated in not having any Linux guys on-site
>>for test #2 -- there wasn't much they could have done.
>
> Hmm.
>
>>It took the c't benchmark to bring out the reality check,
>>showing the specific cause.
>
> Odd, the c't benchmarks I saw seemed to be everything you are accusing the
>Mindcraft ones of being.
>
> Or are you speaking of a different set of benchmarks? Do you have a link?
There'll be a link on here:
http://www.kegel.com/mindcraft_redux.html
It's the June 1999 c't benchmarks. I'm not saying that
they are perfect, but at least they brought out the NIC's
performance bottleneck, rather than just leaving it a
mystery. That's important to people deciding whether a benchmark
is relevent.
[I would like to see better dynamic benchmarks, but it may never be
satisfactory to everyone. NT likes threaded ASP-type stuff, while
the equivalent preferred stuff on Apache would be through php or perl
modules in Apache.]
There's a comparison on the Kegel site above comparing the
three benchmarks.
For example:
The graph where Apache's performance dies is here:
http://www.mindcraft.com/whitepapers/nts4rhlinux-webrps.gif
(from the original paper:
http://www.mindcraft.com/whitepapers/first-nts4rhlinux.html )
Their NewThink comparison, in which they say they are proven correct
in their benchmarking, is here:
http://www.mindcraft.com/whitepapers/ob1-wb-p123.gif
Goodbye.
--
Grant Fischer (gfischer at hub.org)
Steve Sheldon
>This is standard operating procedure in most industries -- choosing the
>correct tool for the job. It is pathetic that this industry has so many
>unqualified people who, by dint of passing a few idiotic tests, think
>they're God's gift to Information Technology. Hate to tell you, guy, but
>there's more to being an IT specialist than knowing how to tweak obscure
>registry settings using "regedit". Have you ever even TAKEN an
>Operations Research course?!
You seem to have this attitude that if people don't use Linux somehow they
aren't thinking. I'm afraid I have quite the opposite opinion, people who
are using Linux are quite likely not thinking about the long term... seeing
a short term cost savings in favor of long term pain.
And my opinion comes from several years of experience... 4 years dedicated
to Unix, 3 years to NT. I took the idiotic tests because I was tired of
arguing with moronic IT managers whether you needed an MCSE or not. Simply
went in one week and took the tests to overcome the obstacle.
I put it in my .sig because it bugs Linux zealots like yourself.
>> Not sure I'd use Linux as a firewall. I value my data too much for that.
>See what I mean? Rigidity. Dismissing a possible solution without
>testing it first to see if it will do the job. I'd fire you in a minute.
Ha, that's ok... You wouldn't even make it past the first interview.
I'm dismissing the possible solution because I've already seen it
implemented numerous time and fail. In each and every case the company
scrapped Linux and went with a professional firewall solution.
>Things have changed, but some old Cobol-heads still insist that Windows
>NT and Visual BASIC are the solution to all problems. (Whoops, that
>should have been "MVS/CICS and COBOL", me bad!).
Huh? Cobol is a mainframe thing. I haven't met any mainframe support
staff who feel that NT or Unix is anything but a toy.
I don't know where you come up with these things...
Steve Sheldon
>The 4-NIC benchmark might be relevent to some large file
>servers. I can't see a typical small office using such a machine.
Uhh... what's a small office have to do with this?
The purpose was to show what happened on data center sized hardware... i.e.
a file server for a company with 3,000 employees.
>I look for benchmarks to show the strengths and weaknesses
>of various solutions. The cynic in me sees that Mindcraft tried
>to give the appearance of varying the configuration, but
>not on the number of NIC's, which just so happened to
>be the key to having NT perform better.
Yes, I'm waiting for the black helicopters as well.
>> And this was attributed to a bug in the kernel code as I recall.
>Yes, very good point. Sort of something that you might want
>to see if there's a work-around, eh?
The benchmark brought this to the attention of the kernel developers.
>>>help, and not succeeding. They initially denied running
>>
>> Which were shown to be true.
>I didn't mean that they tried to ask for help and couldn't get
>the words out. I meant they claimed that they didn't
>receive any offers of help, which is clearly wrong.
We have evidence of one offer... "Linux sucks, use FreeBSD!"
>[Don't be obtuse. They claimed that they didn't get any help,
>but didn't respond to the person offering advice on the
>Internet, and tried to call the RedHat help line for
>benchmark assistance. They should have approached RH
>differently.]
Yes I know, RedHat claimed they should have called Marketing.
>[I would like to see better dynamic benchmarks, but it may never be
>satisfactory to everyone. NT likes threaded ASP-type stuff, while
>the equivalent preferred stuff on Apache would be through php or perl
>modules in Apache.]
It would be curious to see such benchmarks as well. Unfortunately I'm
afraid that when ASP the turns out to be faster there will be great
complaints that they didn't use the right solution for something.
The thing you need to understand is that ASP is nothing more than glue to
hold COM calls together. The core logic doesn't run as VBScript or JScript,
it runs as compiled C++ or VB code.
anyway...
Steve Sheldon
>Steve Sheldon wrote:
>> Anyway, we are not talking about ISP's here. We are talking about your
>> desire to take over the corporate IT world with Linux by replacing Windows
>> NT and Novell Netware. Let's stick to the topic, shall we?
>Excellent, dear sir! Excellent! Attributing to me ideas that I have
>never expressed, while easily dismissing an application where Linux has
>clearly shown itself to be useful (hmm, http://www.deja.com is another
>example). My hat off to you.
Wait a minute. Why did you cut out the comments about Solaris being the
most commonly used solution for ISP's?
>As I'll repeat: The only time to use Linux is when it's the most
>cost-effective solution to a particular task. The same applies to
>Windows NT, Novell, and whatever else you want to consider. I have even
>STATED, in real WORDS, that there are applications where NT is the
>preferred solution -- such as when you have a network to saturate with
>static content from four network cards (grin).
As I recall you made some quirp about it being suitable for small servers.
I've been trying to figure out if a file server handling 2,000 user
connections is small or large.
>Which I have repeated over and over ad nauseum for the past forty
>messages that I have addressed to you.
Awww, now you know how I feel arguing with a brick wall.
>Hmm, I'm going to have to add another FUD Distraction Method, "the straw
>man". Which is where an artificial argument attributed to the
>competition is set up (the artificial argument being that "Linux should
>replace NT in all applications"), and then demolished.
Hmm, is this anything like your comments about MCSE certification?
First you made some puzzling comment about it being expensive, which you
never really fully explained. And then there was the puzzling comment about
how it was some how proof of my God like abilities.
I never quite figured out how that had any bearing on the conversation.
Seemed to be a diversionary tactic.
>Of course, I'm not made out of straw, and I'm not a typical idiot like
>that Matthew person that you keep making a fool out of, so try to be a
>little more inventive in the future.
Oh, I admit you're much smarter than Zsolt and Matthew.
But your constant twisting of the subject matter is most puzzling.
Such as this insistence that we keep talking about internet service
providers. The post I followed up to went on and on about how Linux was
used at most of them... well ok, sure if you only count small ones, but NT
wasn't.
Is this what you call a straw man? It sounds plausible.
Now can you respond to the topic, or are you going to twist it around
again?
Steve Sheldon
>On Sat, 14 Aug 1999 01:46:28 GMT, Steve Sheldon <she...@visi.com> wrote:
>> You still just don't seem to get it.
>>
>> We *HAVE* servers configured very similarly to the ones Mindcraft tested
>>here in our data center. Except we use Compaq, not Dell, and only two NICs
>>not four, except we use gigabit over fibre not 100baseT over cat-5.
>>
>> Why do you keep claiming otherwise?
>>
>> This is what Mindcraft says about the issue:
>>
>>---
>>Why did you use a four-processor server?
> That's not the point. The usage pattern is. Also, it
> wasn't the four processors (as C't demonstrated) that
> were the problem but the 4 NIC's.
Oh my! More spin...
You're a slipperly little critter. At every turn you have a completely new
excuse.
--
Stuart Fox
Grant Fischer <gfis...@amblehub.amble.net> wrote in message
news:slrn7r9jve....@amblehub.amble.net...
> On Fri, 13 Aug 1999 22:21:39 GMT, Steve Sheldon <she...@visi.com> wrote:
> >gfische...@hub.org (Grant Fischer) writes:
> >
> >>>http://www.mindcraft.com/whitepapers/openbench1.html
> >
> >>I always found it funny that one of the key requirements was
> >>that the machine have 4 NIC's. They were willing to experiment
> >>on software, # CPU's, etc, but never the variable that was
> >>bringing out the specific problem in the Linux kernel.
> >
> > It's not an uncommon configuration. Also the goal was to test the
> >throughput of the infrastructure and as such you need suitable
> >infrastructure to push.
>
> Does that make it a common configuration? What perentage of
> servers do you run with 4 NIC's?
You're right, it's not common. I can think of only four that I deal with.
Three are Firewall-1 on NT, with multiple DMZ's (actually one has only one
physical NIC, but has about 16 virtual NICs - using a VLAN card & switch).
The other is runing four NIC's each teamed (so essentially two pairs of two)
and using FEC - the throughput is amazing - this is for an internal LAN
though - it's hard to get an Internet pipe big enough to saturate with 4
NICs...
Chad
Eric Lee Green <e_l_...@hotmail.com> wrote in message
news:CEDE04B5C1CFFF73.683114A2...@lp.airnews.net...
> Chad wrote:
> > Heh.. typical. Linux advocates can't come up with real numbers, so they
must
> > mock
> > all scientific benchmarks conducted by linux "experts" themselves.
>
> Typical. NT advocates can't win if it's turned into a question of "what
> is the most cost-effective platform that performs a particular task"
> (such as, say, departmental firewall), so they turn pointy-haired boss
> on us and throw out benchmarks that are applicable only to some other
> task.
Again, spewing forth the narrowmindedness and shortsightedness...
cost-effective implies TOC. TOC of linux in most configurations is much
higher
seeing as how it has to be nursed along and requires a full-time admin just
to
keep the linux box running.
>
> Repeat after me: Those who repeatedly choose a more expensive solution
> when a less expensive solution will do are doomed to eventual
> unemployment. If your commute needs are served by a Toyota, it is
> irrational to buy a Mercedes. Yeah, it's good ego-boo to say "Yeah, I
> have a Mercedes", but in a business environment it's all about dollars
> and cents, not ego-boo.
More like Yugo and Mercedes. Yugo may cost less, but lasts half as long and
requires
frequent repairs and more money over time.
>
> Right now, Linux is a Toyota, and NT is an SUV. NT will do some things
> that Linux won't do, but Linux is cheaper and sips less gas (runs on
> smaller computers). If you need four network cards and four processors,
> use NT. If you need a firewall, use Linux. Most cost-effective solution
> for a particular task, no?
need a firewall use linux? hah! I wouldn't trust a one-node network to
protection
from a linux firewall (or firecheesecloth would be more analogous)
The "best" firewalls are either hardware based or software based on NT
(Raptor, FireWall-1, etc)
or some variation of BSD.
>
> When you're fired because you continue to insist on using a
> more-expensive solution to a particular problem when a less-expensive
> solution exists, don't bother asking me for a job (even though we're
> hiring). I'm not interested in inflexible poltroons who are more
> interested in egoboo than in getting the work done in the most
> cost-effective manner.
How 'bout when you're fired because you had to explain to the CEO
that your site went down or got hacked because you wanted to
save a few bucks up front ?
>
Chad
We can argue all day about wether or not someone _really_ needs 4 NICs.
The fact of the matter is, many people probably do, or will knowing that
NT's
performance can be significantly increased by adding a few more NICs.
On the other hand, they have to rely on 3rd party solutions or dump linux
all together to get the performance they need.
<sarcasm> But hey, Linux is ready for the enterprise! </sarcasm>
You just admitted it didn't scale well. Period, end of discussion.
Chad
Grant Fischer <gfis...@amblehub.amble.net> wrote in message
news:slrn7r9jve....@amblehub.amble.net...
Chris Costello
>> : What is a brain surgeon doing on USENET?
>>
>> Ask Hobbyist... *grin*
> OK, OK, I know that my type is uncommon around these parts. A few
> of us do have more than average interest in computing.
Mistyped anyway.
I meant to type
`What is a brain surgeon doing on advocacy groups in
USENET?'
--
|Chris Costello
|It wasn't as easy to get programs right as we had thought. - Wilkes, 1949
`--------------------------------------------------------------------------
Grant Fischer
>gfis...@amblehub.amble.net (Grant Fischer) writes:
>
>>The 4-NIC benchmark might be relevent to some large file
>>servers. I can't see a typical small office using such a machine.
>
They are also readers of benchmarks, and need to see if one
is applicable to their purposes. Excluding sensitivity testing
on the NIC's doesn't give a complete picture, and MS's
comparison pages doesn't make it clear how well Linux
performs against NT at the low end.
> The purpose was to show what happened on data center sized hardware... i.e.
>a file server for a company with 3,000 employees.
Trouble is, they don't seem to make this clear by providing a
range of benchmarks to cover the low end. As many sane IT types
will tell you, use the best tool for the job. For many, this
is be a range of Linux/UNIX style solutions -- Linux is
attractive for the low end, and it migrates well up to the
high end UNIX's.
For other places, NT is the way to go. Just think about the
choices in a rational manner.
>>I look for benchmarks to show the strengths and weaknesses
>>of various solutions. The cynic in me sees that Mindcraft tried
>>to give the appearance of varying the configuration, but
>>not on the number of NIC's, which just so happened to
>>be the key to having NT perform better.
>
> Yes, I'm waiting for the black helicopters as well.
Funny. I said the cynic in me, but this highlights the problem
with a vendor-sponsored benchmark. They aren't being paid to
come up with useful results, but rather some result the vendor
can tout. In this case, it means ignore the low end totally
(that would be inconvenient to have show up on the same
press release) and focus on a very precise configuration
(which some say was only fixed in the latest service packs.)
>>> And this was attributed to a bug in the kernel code as I recall.
>
>>Yes, very good point. Sort of something that you might want
>>to see if there's a work-around, eh?
>
> The benchmark brought this to the attention of the kernel developers.
I hardly think they would have let NT's performance stand had
it dropped off like that. There would have been major geek
sweat expended.
>
>>>>help, and not succeeding. They initially denied running
>>>
>>> Which were shown to be true.
>
>>I didn't mean that they tried to ask for help and couldn't get
>>the words out. I meant they claimed that they didn't
>>receive any offers of help, which is clearly wrong.
>
> We have evidence of one offer... "Linux sucks, use FreeBSD!"
Where *do* you get your information? Does it come in little
pamphlets, or flashes of divine inspiration?
http://x26.deja.com/[ST_rn=ap]/getdoc.xp?AN=453900037&
search=thread&CONTEXT=934603047.635568184&hitnum=11
(URL broken up to keep under the line length convention)
>
>>[Don't be obtuse. They claimed that they didn't get any help,
>>but didn't respond to the person offering advice on the
>>Internet, and tried to call the RedHat help line for
>>benchmark assistance. They should have approached RH
>>differently.]
>
> Yes I know, RedHat claimed they should have called Marketing.
Well, yeah. If I've ever had to do any performance testing,
I talk to the sales people in order that they will get me
into the proper technical gurus. The sales people are motivated,
and know who to talk to. I've been in some situations where
we were prepared to go to California for a week to benchmark
an app on a vendor's large server. The idea was to add CPU's
in groups of five to see roughly where we needed to be.
This isn't something I'd ask the tech support line about.
>
>>[I would like to see better dynamic benchmarks, but it may never be
>>satisfactory to everyone. NT likes threaded ASP-type stuff, while
>>the equivalent preferred stuff on Apache would be through php or perl
>>modules in Apache.]
>
> It would be curious to see such benchmarks as well. Unfortunately I'm
>afraid that when ASP the turns out to be faster there will be great
>complaints that they didn't use the right solution for something.
With such an unbiased view, I can see where you are coming from.
[You should say "if ASP the turns out to be faster."]
> The thing you need to understand is that ASP is nothing more than glue to
>hold COM calls together. The core logic doesn't run as VBScript or JScript,
>it runs as compiled C++ or VB code.
I know.
People are saying that suggesting that php or perl as apache modules
have a very good chance of keeping up to such an approach. The modules
mean that the perl interpreter is alive, and has preparsed the program.
A good programmer will keep persistent database connections through
mod_perl techniques, saving a lot of construct/destruct overhead.
The techniques in perl are simple and straightforward. A good
perl programmer is very productive in that language. Even if
the pure perl stuff was about 80% as fast as ASP, there's a lot
of benefit to going that route.
And if an ASP approach still beats that, you can always extend Perl
using native C/C++/etc extensions to do any complex stuff that doesn't
execute fast in Perl.
If you still need a boost, you can execute the Perl stuff offline,
pregenerate common pages and put them in place to be served statically.
Generate them every "n" minutes or so, and let the front-end proxies
do their thing.
And if an ASP approach still beats that, you can always extend Apache
itself by writing a module.
Options are a good thing. And these are just the tricks I can think
of, and I don't run a big web server.
Andy Newman
X-Mailer: Mozilla 4.6 [en] (X11; I; FreeBSD 3.2-RELEASE i386)
Sensible lad.
Andy Newman
> that is not based upon one of the well-known embedded Internet operating
> systems, and I seem to recall quite a few security exploits of their
> PIX firewall.
Lucent have one based on Inferno.
John Hughes
> > > As a departmental file and print server, yes. As a small business or
> > > small government database server, yes (city of Garden Grove CA, half
the
> > > school districts in Louisiana, Enhanced Software Technologies Inc.).
As
> > > a web server, EMAIL server, and internet server, yes (see
> > > http://www.deja.com and about 90% of the Internet Service Providers
out
> > > there).
> >
> > Bollocks. Get a list of ISP's and do the netcraft test on them. Its all
FUD.
> >
> > We did the test in the UK and a very small amount were using Linux.
>
> island with delusions of grandeur where they think boiling food is the
> best way to cook it and where there are few ISP's because of a state
> monopoly phone system that insists upon charging per-minute fees for
> Internet access. We long ago figured out here in the 'States that state
> monopolies for phone service were a bad idea, that's why we have over 40
> million people connected to the Internet and why your damp foggy island
> has, what, 8 people connected to the Internet? (heheh).
>
> Brief survey of the ISP's in Shreveport, Louisiana, a typical mid-sized
> city in the United States, finds a total of 18 ISP's, of which 12 are
> using Linux. The others are using BSDI or Solaris, by the way, so don't
> get your hopes up about NT. In fact, not a single major ISP in the
> U.S. West? No. IBM? No. Mindspring? No. If you can name one (1) major
> ISP (as in, available in at least two states if not nationwide) that
> uses Windows NT, I'll happily mail you a Cajun Cooking cookbook and
> assortment of spices so that you can cook some real food (grin). (Note
> -- no boiling involved, though, so you may have to figure out how to
> saute' and carmelize, but I'm sure you can do it!).
>
>
> --
> Eric Lee Green http://members.tripod.com/e_l_green
> mail: e_l_...@hotmail.com
> ^^^^^^^ Burdening Microsoft with SPAM!
Boiling???!!!! What you talking about.......? Seen your shrink lately?
OK, so in the UK we arent using Linux as much in the US. Im not advocating
NT for ISP's and never did. So dont put words in my mouth.
John Hughes
Steve Sheldon <she...@visi.com> wrote in message
news:d74t3.3453$Qn4.3...@ptah.visi.com...
> >> Bollocks. Get a list of ISP's and do the netcraft test on them. Its all
FUD.
> >>
> >> We did the test in the UK and a very small amount were using Linux.
>
> >city in the United States, finds a total of 18 ISP's, of which 12 are
> >using Linux. The others are using BSDI or Solaris, by the way, so don't
> >get your hopes up about NT. In fact, not a single major ISP in the
>
> different than the corporate IT world which NT was targetted for.
>
> As far as number of ISP's using Linux. If you do a weighted average on
> that... i.e. weight the ISP by the number of customers you will find that
> Linux is in fact not used by any of the major players.
>
> An ISP with 500 customers is quite different from one with 15,000 would
you
> not agree? Shouldn't the ISP with 15,000 customers count as being more
> important than the one with 500?
>
>
> desire to take over the corporate IT world with Linux by replacing Windows
> NT and Novell Netware. Let's stick to the topic, shall we?
> Steve Sheldon email: she...@yuck.net
> BSCS/MCSE url: http://www.sheldon.visi.com
> Save a flag... Burn the Constitution.
The reason this went to ISP's is that the Linux zealots think that they
already have taken over the ISP world. This relevant is it not?
John Hughes
<je...@dementia.mishnet> wrote in message
news:slrn7r941...@dementia.mishnet...
> On Fri, 13 Aug 1999 18:56:50 +0100, John Hughes <ma...@easynet.co.uk>
wrote:
> >
> >Grant Fischer <gfische...@hub.org> wrote in message
> >news:slrn7r87ha.ci3....@dogbert.cibg.tdbank.ca...
> >> On Fri, 13 Aug 1999 04:15:54 GMT, Steve Sheldon <she...@visi.com>
wrote:
> >> >je...@dementia.mishnet () writes:
> >> >
> >> >> Hardly. What we complained about was that the test was
> >> >> a joke bought by MS. The subsequent and previous benchmarks
> >> >> by both Mindcraft and others demonstrated that.
> >> >
> >> > Oh, in case you missed your dose of reality check...
> >> >
> >> >http://www.mindcraft.com/whitepapers/openbench1.html
> >> >
> >>
> >> I always found it funny that one of the key requirements was
> >> that the machine have 4 NIC's. They were willing to experiment
> >> on software, # CPU's, etc, but never the variable that was
> >> bringing out the specific problem in the Linux kernel.
> >>
> >> caused so much fuss, had the Linux performance collapsing
> >> diastrously. They also made claims about trying to seek
> >> be implemented before benchmarking -- they wanted those
> >> simple changes ruled out as the cause of the bad performance.
> >> In other words, at least give it a fair chance.
> >>
> >> despite how much they'd like us to forget it.
> >>
> >> enough together to show that there wasn't much the Linux
> >> guys could do without fixing the network stack lock.
> >> Mindcraft was vindicated in not having any Linux guys on-site
> >> for test #2 -- there wasn't much they could have done.
> >>
> >> showing the specific cause.
> >>
> >>
> >> Grant Fischer (gfischer at the domain hub.org)
> >>
> >
> >Cmon the C't benchmark was just as unfair as the Mindcraft #1.
> >
> >CGI on IIS isnt an option as you have more scalable options.
>
> 'cgi' isn't the only option on Linux either.
>
> >
> >To this date I havent saw a benchmark which shows Linux scaling with
todays
> >technologies.
>
> Ditto for NT. That's one of the 'interesting' bits about Mindcraft.
> It doesn't bear any resemblance to sites that serve that kind of
> volume.
>
In what way?
Would you agree that NT scales better than Linux?
John Hughes
Geoff Winans <gwi...@campuscw.net> wrote in message
news:37b4c813....@news.campuscw.net...
>
> >
> >Bollocks. Get a list of ISP's and do the netcraft test on them. Its all
FUD.
> >
> >We did the test in the UK and a very small amount were using Linux.
>
>
> Quit reading so deep into things. You make yourself look anal that
> way.
>
>
> ~Geoff
I quote :
> > Yet linux is ready for the enterprise *choke* yeah, sure.
>
> As a departmental file and print server, yes. As a small business or
> small government database server, yes (city of Garden Grove CA, half the
> school districts in Louisiana, Enhanced Software Technologies Inc.). As
> a web server, EMAIL server, and internet server, yes (see
> http://www.deja.com and about 90% of the Internet Service Providers out
> there).
Please dont filter out the reality!