Opinions wanted: HP vs SGI vs SUN
Rupert Thurner
>
> In article <3hrqmi$d...@pith.uoregon.edu> b...@chinook.uoregon.edu (Ben Marcotte) writes:
>
> I haven't seen much mention of DEC here yet. Which brings me to a point,
> why are so many people religiously attached to Sun's? In my current
> search for a new workstation, they rank at the bottom of my list for
> performance and OS reliability. I have not yet gotten to play with a Dec
> Alpha yet, but from the numbers and the discussion about them in the
> various user support newsgroups, I would currently rank DEC first.
>
> I have admined every major workstation (Sun3, Sun4, SGI, DEC, HP, IBM,
> DEC Alpha, DEC Station, Linux, SCO Unix) and let me tell you DEC ranks
> real low on my list. The recent release of OSF1 V3.0 is a licencing
> nightmare. Only two users are allowed to login at once (root does not
> count) under the default licence. A licence has to be purchased for
> each additional user. These licences are yearly, so good paperwork
> has to be kept on the machine. Then, the little quirks of keeping
> them running.... I have a DEC alpha loaded with 3.5 internal drives,
> and it has a heat problem. This was a suprise to the DEC Engineers,
> and they only offered advice was move it to a colder room or get some
> of the drives out. We also run some pretty intensive programs on the
> DEC, and a single STAGS (Statistical Analysis Program) can bring the
> machine to its knees, with nothing else running (above system
> software). The DEC support people have alot to learn about customer
> support!! The ONLY DEC persons who seemed to want to help when I
> called was a Licencing person (Chris Kelley) who helped me through my
> nightmare of upgrading licences.
>
> As far as HP, SGI, or IBM (no one seems to have mentioned
> RS/6000's either), the race for second is too close to call. HP's and
> IBM's have higher floating point performance. SGI's have better graphics
> performance (of course). None of these systems seems to have a large
> advantage over each other in price, but all have much higher
> price/performance ratios than the Alpha's.
>
> If you want a really good price/performance ratio, I would
> suggest a Pentium based machine. A P5-100 pulls in SPEC numbers better
> than the entire Sparcstation 5 line and is comparable with the 150Mhz
> Indy, the Sparcstation 20 model 51, or the RS/6000 models C10 and 41T.
> And there are more OS's to choose from (WinNT, many SVR4's, many bsd's).
>
> I would appreciate hearing anymore that anyone has to say on this subject.
>
> I think a major issue you are not seeing is the man hours to support a
> machine. A lot of people like Suns because the Administration of them
> is rather easy. SGIs, a little more difficult (IMHO), but I have
> found that the customer support bends over backwards. I had problems
> with the IRIX 6.0.1 FORTRAN compiler, and within a few hours, I had
> one of the Developing programmers helping me with my problem. He
> found the bug in the compiler, and gave me a work around, I lost all
> of 4 hours work on this problem.
>
> My bottom line is this, I support several different arcitectures, as
> well as a testing facility, so my admin time is limited when we are
> under full scale testing, so I want machines that have proven
> reliability. For me, Suns and SGIs are at the top of the list....
> You really have to equate supporting the machines in any
> price/performace ratio you look at. (again IMHO)...
>
> Wayne
what do admins of other systems think of this?
rupert
Butch Deal
Suns with Solaris 2.x are about the easyest to maintain, configure, install,
upgrade, etc, etc. and are about the most reliable.
SGI's with IRIX 5.3 are not too bad. pretty reliable although I have been
dealing with a bad space ball on one fore months (the sgi tech. has replaced
almost the entire system and reinstalled several times). They come by default
with a strange one partition disk and repartitioning is not all that easy. I
would have to rate then higher than HP's and AIX systems though. DEC's don't
really count in the comparison till they get a little closer to UNIX.
--
#include <std/*>
The Butcher
Butch Deal de...@ait.nrl.navy.mil
--------------------------------------------------------------------------------
Martin Knoblauch
S. Woan) writes:
|>
|> I guess some of you haven't seen our Visual System Management tools
|> that ship standard on AIX 4.1 and as a RPQ for AIX 3.2.*, if I
|> remember correctly. Real cool and makes lots of common system admin
|> stuff a breeze especially for people from the PC world, but only for
|> AIX.
|> --
And of course you should not forget the absolutely amazing
IndigoMagic admin tools for IRIX-5. I speak from an totally
unbiased point of view, of course :-)
Martin
--
+---------------------------------+-----------------------------------+
|Martin Knoblauch | Silicon Graphics GmbH |
|Application Center | Am Hochacker 3 - Technopark |
|Silicon Graphics Computer Systems| D-85630 Grasbrunn-Neukeferloh, FRG|
| | Phone: (+int) 89 46108-179 or -0 |
| | Fax: (+int) 89 46108-190 (-222) |
+---------------------------------+-----------------------------------+
|Network: <kn...@munich.sgi.com> | V-Mail: 5-8935 | M/S: IMU-315 |
+---------------------------------------------------------------------+
Frank Kraemer
|> And of course you should not forget the absolutely amazing
|> IndigoMagic admin tools for IRIX-5. I speak from an totally
|> unbiased point of view, of course :-)
it is so *Magic* that after a failed -> build new kernel -> reboot
procedure -> build new kernel -> reboot.....the sceen is magic black
(guess I made a mistake - that's for shure)
Rebooting to make some small changes valid is early 1970 Unix....
.....but indeed it is faster than AIX rebooting !!
-frank-
+----------------------------------------------------------------------+
| E-Mail : krae...@vnet.ibm.com / kr...@ibm.de |
| Voice : +49-(0)611/776-364 / Fax: +49-(0)611/776-500 |
| Mail : Abraham-Lincoln Street 26, 65189 Wiesbaden, Germany |
+----------------------------------------------------------------------+
Steve Strange
> would have to rate then higher than HP's and AIX systems though. DEC's don't
> really count in the comparison till they get a little closer to UNIX.
Huh? Please provide your definition of UNIX.
Steve
--
Steve Strange
Digital Equipment Corporation, Nashua, NH
DCE DFS for DEC OSF/1 -- Development
str...@zk3.dec.com
Ronald P. Larkin
This thread is most useful to someone who needs unix boxes, has no unix
experience yet, and needs perspective. Thanks especially to the
administrators of diverse sites who have compared different systems
directly! Is there any other source of such informed opinion? Perhaps
even the massive crossposting is justified....
Ron Larkin
--
Ron Larkin (r-la...@uiuc.edu)
Illinois Natural History Survey, Champaign IL 61820
Will Morse
I have joined this thread a little late, but I would add Clones. The
only clones I have worked with are Sun clones from Tatung (which are
better than Suns, IMHO) but Tatung also makes a PREP box which is
basically an RS/6000 clone. I don't know of any HP or SGI clones,
and no one in their right mind would make an Alpha clone.
Will
--
# Gravity, # Will Morse
# not just a good idea, # BHP Petroleum (Americas) Inc.
# it's the law. # Houston, Texas
# # wi...@starbase.neosoft.com
#
# These are my views and do not necessarly reflect the views of BHP !
Nathan F. Janette
writes:
>
> We received a "server" from HP, supposedly "just what we needed" to be our
> mail-news-web server...it a HP9000/800 G30 (847) with 512Meg of RAM and 2
RAID
> arrays each on their own Fast-Wide SCSI-2 bus...
>
> I am completely disappointed with it. I've ran countless benchmarks on the
> drives in all 3 "supported", and some non-supported modes, and the best I can
> muster is around 750K/sec writes and 1.6M/sec reads. Our SS20 with a single
> drive can double that. On top of that the current HP-UX allows only 4G
> partitions, so we can't have a single 8G partition for news...and it sure is
> great to have all that RAM...I increased the number of I/O buffers, but now
> when sync comes around to dump them, the machine basically locks up for about
> 30secs...pop mail clients time-out trying to connect...not so cool.
It is the RAID subsystem that is slow, not the disks.
Such performance is the cost of using many RAID
systems. It's obviously a trade-off for good data
security.
We use an HP/apollo 9000-735/99 for our NFS server.
It has 144 MB RAM and 14 FWD disks (2-4 Seagate
Barracuda disks, 30 GB total). With the write cache
enabled, we can write to a disk at around 7 MB/sec,
and read a bit slower(!).
We also experienced the sync-wedges, and just
installed a patch that limits the RAM-caching to 1/2
the RAM amount, which has helped solve the problem.
The biggest problem for us is that HPUX NFS seems to
have major problems with tar usage (mentioned briefly
in a post by someone at HP a few weeks ago). When I
try to install a NEXTSTEP package (which uses tar)
from a NEXTSTEP system onto the /LocalApps area, which
is served via NFS from the HPUX system, it fails. The
problem is that ownership of the extracted files isn't
changed to the user, which becomes a show-stopped when
a directory is involved. This is pathetic behavior,
and I hope a serious embarrassment to the network
engineers at HP. A fix *must* be implemented ASAP by
HP.
--
Nathan Janette
Systems Manager, Axel T. Br nger Lab
Internet: nat...@laplace.csb.yale.edu
Voice: 203 432 5065
Fax: 203 432 3923
Ben Marcotte
: In article <3hvfrk$c...@osiris.wu-wien.ac.at>, Rupert Thurner <rthu...@x11srv.edvz.uni-klu.ac.at> writes:
: |> wa...@suni202.larc.nasa.gov (C Wayne Huling) wrote:
: |> what do admins of other systems think of this?
: |>
: |> rupert
: sounds about right here.
: Suns with Solaris 2.x are about the easyest to maintain, configure, install,
: upgrade, etc, etc. and are about the most reliable.
(cough cough sputter sputter - pick myself up off the floor)
What color is the sky on your planet?
Solaris - easy, reliable? You can use those three words in one sentence?
It has been the biggest pain in the *ss for us!
For me it constantly feels like Sun is working too hard on the toys for the
OS (Wabi, SDK's, etc...) and not enough on the basics. By basics I mean:
1) Sol 2.1 through 2.3 can't keep utmp[x] records straight. Try
'finger -l' on someone and you'll see all their previous logins (that
have been logedout) on any ttys (ptys) that someone else has not used yet.
I have heard that 2.4 has the same bug.
2) The lp services have a number of problems. The
/var/spool/lp/system/pstatus file is not correctly formatted. By default,
the banner page is turned on _AND_ turning it off in admintool doesn't
turn it off _AND_ Sun tech support couldn't even tell me how to turn it
off. It wasn't until someone on comp.unix.solaris was complaining about
the same thing and got a response (finally) that we found a solution (you
have edit the /etc/lp/interfaces/[printername] file by hand and change
nobanner="no" to "yes"). We recently had our printer power cycle on us
which should not have caused too many problems for a fault tolerant OS, but
instead Solaris let the device drivers in the kernel get corrupted and we
were forced to reboot the system (and, no, restarting the printer daemons
did absolutly nothing).
3) Have you followed any of the threads in comp.unix.solaris on setting
up a modem for dial-in? I challenge you to find anyone on that group who
will say that it is easy to "maintain, configure, install, and upgrade" a
bidirectional modem on a Sparc. I've been trying for months to get ours to
work under 2.3 and I have not been able to get anything out of it.
4) Lets not forget that Solaris 2.1 was shipped with a completely unusable
copy of xdm. I call that releasing your beta copy NOT releasing a
"reliable" product.
There are many more problems that a quick glance through the sun
newsgroups would reveal.
: SGI's with IRIX 5.3 are not too bad. pretty reliable although I have been
: dealing with a bad space ball on one fore months (the sgi tech. has replaced
: almost the entire system and reinstalled several times). They come by default
: with a strange one partition disk and repartitioning is not all that easy. I
: would have to rate then higher than HP's and AIX systems though. DEC's don't
: really count in the comparison till they get a little closer to UNIX.
^^^^^^^^^^^^^^^^^^^^^^^
Care to explain why OSF/1 does not meet you standards but an OS with the
problems I listed above does? I tend to think that a system that can't
even keep track of who is logged in correctly is not a complete Unix!
: --
: #include <std/*>
: The Butcher
: Butch Deal de...@ait.nrl.navy.mil
: --------------------------------------------------------------------------------
==============================================================================
Ben Marcotte b...@chinook.uoregon.edu (503) 346-4592
==============================================================================
Michael Rogero Brown (Sys Admin)
: > would have to rate then higher than HP's and AIX systems though. DEC's don't
: > really count in the comparison till they get a little closer to UNIX.
: Huh? Please provide your definition of UNIX.
Yes, that a bit of a strange comment. Some would say OSF/1 (which DEC is
using instead of Ultrix) is more Unix then AIX.
I hope this is not another "if it's not SVR4, it's not Unix" type argument.
Personally if its based on either SVRx or BSD, its Unix in my book. [as if
I decide what;s Unix or not. :)]
Spec 1170 is now supposed to be the determining factor of "unix or not", but
when I hear things like IBM's plans on getting OS/400 to pass 1170, it makes
me wonder. [don't know anything about os/400, except that I wouldn't think it
Unix]
--
----------All Opinions Expressed are MINE, not IBM's--------------
Michael Rogero Brown (uK Development System Administrator)
IBM (uK Development) TEL/TIE (407) 443-6400
Boca Raton, FL Internet: mi...@bocaraton.ibm.com
If you think I speak for IBM, then I've got some swamp land^H^H^H^H^H^H^H^H^H^H
real estate to sell you.
Mike Suzio
I work in a facility with just about every workstation configuration under
the sun (no pun intended) - Sun (Sun OS and Solaris), SGI, HP, DEC Alpha,
IBM, Cray, Sequent, etc. In the past, I've worked extensively with SGI's
and Suns in smaller support environments (less than 20 users).
I'd say that for small groups without a lot of time to spend on system
administration, SGI's are the best. They have a lot of "point and click"
administrative tools that make a sysadmins life easier, and very good
on-line documentation. It's not perfect, but I managed to get a group of
scientists setup with a couple of SGI's that they could self-administrate
after I left my sysadmin position - I was able to render myself unneeded
because the machines were pretty much no-brainers to administrate.
SGI is also very good on the "power" scale - the machines are fast (I/O
subsystem speed almost always helps, no matter what you do), and the
Challenge/Onyx server line very nicely integrates with the smaller desktop
systems.
HP machines also come with a very good sysadmin GUI tool (sam). Overall,
the machines are way complicated (IMHO). I wouldn't recommend them for
a novice sysadmin. They are decently powerful machines with good graphics,
however (my department uses mostly HP machines).
Sun's are easy to administrate for more experienced admins (it takes a
while to be a good Sun geek, but after you get the basics down it's
smooth sailing). As far as I know, no tools are shipped with Suns like
you get with SGI's or HP's. Still a very solid system however, and I think
Sun offers very attractive pricing to educational customers (which is why
they seem to dominate the college market).
DEC's are pretty low on my list, and are a marginal machine type in our
company. They are fast, no doubt - but the OS is yucky to the extreme (DEC
OSF 3.0 is the worst of OSF/1 with some extra stuff DEC seems to think is
nice), the support is unresponsive, and it's a bitch to get most of our
software (CAE applications) to compile and run cleanly on the machines.
Hard to find third-party support for DEC machines, too (whereas we can get
most any kind of software for the other platforms we support).
IBM machines are OK - I have nothing good or bad to say about them. They
don't exactly give me a thrill, however.
- Mike
Ronald S. Woan
I guess some of you haven't seen our Visual System Management tools
that ship standard on AIX 4.1 and as a RPQ for AIX 3.2.*, if I
remember correctly. Real cool and makes lots of common system admin
stuff a breeze especially for people from the PC world, but only for
AIX.
--
+ Ronald S. Woan (IBM VNET)WOAN AT AUSTIN, wo...@exeter.austin.ibm.com +
+ outside of IBM wo...@austin.ibm.com or Compuserve: 73530,2537 +
+ URL: http://cactus.org/~woan/ +
David Mathog
>and no one in their right mind would make an Alpha clone.
Well there must be some crazies out there. You can buy one now from Aspen
Systems and rumor has it that Nekotech is finishing up the firmware that
will let their new systems run it too (possibly also available as an
upgrade to their old systems.) All of these machines will run WNT or OSF/1
or VMS, and if you've got the bucks for three different system disks and
all of those OS's, you can switch between them on the same machine by just
rebooting. Kind of handy for developers who work in all three OS's, they
only need one workstation instead of three.
Regards,
David Mathog
mat...@seqvax.bio.caltech.edu
Manager, sequence analysis facility, biology division, Caltech
Paul Smith
%% whi...@christie.Meakins.McGill.CA (Whitney de Vries) writes:
wdv> AIX's Journaled File System is a big plus. Switching back to
wdv> fragile file systems on other OS's is like taking a walk back
wdv> into the past when unix used to be a Ken Thompson's OS for
wdv> games.
If you want a hardened commercial UNIX filesystem you should check out
the one on DG/UX. Solid.
Plus you can grow 'em, shrink 'em, and chop 'em up, not all of which you
can do with AIX's JFS (at least not 3.4.x, which is what I have). DG/UX
beats AIX hands down in every way, IMHO. Except in popularity,
unfortunately for them ;) :)
--
-------------------------------------------------------------------------------
Paul D. Smith <psm...@wellfleet.com> Network Management Development
Senior Software Engineer Bay Networks, Inc.
-------------------------------------------------------------------------------
"Please remain calm...I may be mad, but I am a professional." --Mad Scientist
-------------------------------------------------------------------------------
These are my opinions--Bay Networks takes no responsibility for them.
Butch Deal
Ben Marcotte <b...@chinook.uoregon.edu> wrote:
>Butch Deal (de...@enterprise.ait.nrl.navy.mil) wrote:
>: In article <3hvfrk$c...@osiris.wu-wien.ac.at>, Rupert Thurner <rthu...@x11srv.edvz.uni-klu.ac.at> writes:
>: |> wa...@suni202.larc.nasa.gov (C Wayne Huling) wrote:
>: |> what do admins of other systems think of this?
>: |>
>: |> rupert
>: sounds about right here.
>: Suns with Solaris 2.x are about the easyest to maintain, configure, install,
>: upgrade, etc, etc. and are about the most reliable.
>
>(cough cough sputter sputter - pick myself up off the floor)
>
>What color is the sky on your planet?
Red..... and yours ?
>
>Solaris - easy, reliable? You can use those three words in one sentence?
>It has been the biggest pain in the *ss for us!
sorry. I can recommend a few good admins.
>
>For me it constantly feels like Sun is working too hard on the toys for the
>OS (Wabi, SDK's, etc...) and not enough on the basics. By basics I mean:
By basics I mean networking, kernel, standards, libs.
>
>1) Sol 2.1 through 2.3 can't keep utmp[x] records straight. Try
>'finger -l' on someone and you'll see all their previous logins (that
>have been logedout) on any ttys (ptys) that someone else has not used yet.
>I have heard that 2.4 has the same bug.
haven't seen that one, but then I have installed 101318-59, 101533-03,
and 101615-02 patches.
>
>2) The lp services have a number of problems. The
>/var/spool/lp/system/pstatus file is not correctly formatted. By default,
>the banner page is turned on _AND_ turning it off in admintool doesn't
>turn it off _AND_ Sun tech support couldn't even tell me how to turn it
>off. It wasn't until someone on comp.unix.solaris was complaining about
>the same thing and got a response (finally) that we found a solution (you
>have edit the /etc/lp/interfaces/[printername] file by hand and change
>nobanner="no" to "yes"). We recently had our printer power cycle on us
>which should not have caused too many problems for a fault tolerant OS, but
>instead Solaris let the device drivers in the kernel get corrupted and we
>were forced to reboot the system (and, no, restarting the printer daemons
>did absolutly nothing).
I have a sparc printer on our main server. Just power-cycled it for
kicks. no problems.
>
>3) Have you followed any of the threads in comp.unix.solaris on setting
>up a modem for dial-in? I challenge you to find anyone on that group who
>will say that it is easy to "maintain, configure, install, and upgrade" a
>bidirectional modem on a Sparc. I've been trying for months to get ours to
>work under 2.3 and I have not been able to get anything out of it.
too many GUI admins on that news group for me.
>
>4) Lets not forget that Solaris 2.1 was shipped with a completely unusable
>copy of xdm. I call that releasing your beta copy NOT releasing a
>"reliable" product.
>
>There are many more problems that a quick glance through the sun
>newsgroups would reveal.
Oh I am sure there are lots....... like why ps -aux craps out.
>
>: SGI's with IRIX 5.3 are not too bad. pretty reliable although I have been
>: dealing with a bad space ball on one fore months (the sgi tech. has replaced
>: almost the entire system and reinstalled several times). They come by default
>: with a strange one partition disk and repartitioning is not all that easy. I
>: would have to rate then higher than HP's and AIX systems though. DEC's don't
>: really count in the comparison till they get a little closer to UNIX.
> ^^^^^^^^^^^^^^^^^^^^^^^
>Care to explain why OSF/1 does not meet you standards but an OS with the
>problems I listed above does? I tend to think that a system that can't
>even keep track of who is logged in correctly is not a complete Unix!
a few patches will get you along way. Course there are those that think no
patches means no bugs, when it really means no solutions.
Ami Varsanyi
environment. Prior to working with SGI's, I worked with SunOS.
Comparing the three, I would have to say SGI's are the easiest
to babysit, Sun coming in second and IBM third. I have been
pleasantly surprised at how robust AIX is after hearing many horror
stories from my peers. But it's "robustness" has led to complexity
that SGI and Sun don't have. And, of course, IBM just HAD to do
some things differently than everyone else.
-Ami Varsanyi
Paranet, Inc.
New Orleans, LA
Art Stine
I have joined this thread a little late, but I would add Clones. The
only clones I have worked with are Sun clones from Tatung (which are
better than Suns, IMHO) but Tatung also makes a PREP box which is
basically an RS/6000 clone. I don't know of any HP or SGI clones,
and no one in their right mind would make an Alpha clone.
There are several companies already doing Alpha 'clones' ... btw, why
do you think "no one in their right mind would make an Alpha clone" ?
-art stine
visix software
John Soward
We received a "server" from HP, supposedly "just what we needed" to be our
mail-news-web server...it a HP9000/800 G30 (847) with 512Meg of RAM and 2 RAID
arrays each on their own Fast-Wide SCSI-2 bus...
I am completely disappointed with it. I've ran countless benchmarks on the
drives in all 3 "supported", and some non-supported modes, and the best I can
muster is around 750K/sec writes and 1.6M/sec reads. Our SS20 with a single
drive can double that. On top of that the current HP-UX allows only 4G
partitions, so we can't have a single 8G partition for news...and it sure is
great to have all that RAM...I increased the number of I/O buffers, but now
when sync comes around to dump them, the machine basically locks up for about
30secs...pop mail clients time-out trying to connect...not so cool.
--
John Soward 'The Midnight Sun will burn you up'
University of Kentucky -the Cure, Piggy in the Mirror.
sow...@pop.uky.edu (NeXT) <a href="http://www.uky.edu/~soward">JpS</a>
Barry Shein
World.std.com (The World) is an SGI Challenge XL with 8x200MHZ R4400s,
512MB, 5xSCSI (3 F+W, 2 F), etc. We also have an SGI CHallenge L with
4x100MHZ, 256MB, 3 SCSI, etc as a news server and a FDDI between them.
We're running IRIX 5.3.
I'm surrounded by various other Suns, BSDI, AIX.
For big machines the SGI is an awesome powerhouse. We still have a few
nits with the FDDI tho found a workaround, they're working on it, but
any problems are down to things like that, the FDDI is in production
so hard to complain too much, it works.
For desktop personal computing at a reasonable price I like BSDI on an
x86 (hard to beat the price), if you need fancy (graphics,
multi-media) then an SGI desktop. But I suppose if you need access to
a lot of general-purpose apps a Sun desktop is still something to
consider.
I used to be a big Sun/SunOS fan, owned a lot of Suns in my day (and a
few Solbourne multis, Sparc, SunOS), but once they went to SVR4 it
didn't seem to matter much compared to something like IRIX (also
SVR4), and it was clear to me Sun was going thru pains changing over
to SVR4 that I didn't want to share that with them. IRIX was about 2
years ahead of them on SVR4 and, more importantly, SVR4/SMP. Maybe
some day when they get it figured out I'll consider a big Sun again, I
think Sun remains an interesting company, it's too bad they managed
the switchover to SVR4 so poorly (IMHO, but follow the money.)
We let DEC bid an Alpha but at the time (late last spring) they didn't
even have a multi in production so that was the end of that, they said
they'd have one soon but we couldn't afford to be an early adopter for
a mission-critical app. I'm still nervous about DEC to be blunt, too
many years of schizophrenia, are they going to dump Unix and run to NT
or something? Or just do both (oh yeah "openVMS" too) poorly and war
internally to the detriment of their customers? I dunno, always seems
to be like that.
I've found SGI support to be good, certainly responsive, mostly
effective even, and I'm a pretty fearsome customer when it comes to
support having run large installations, including Unix, for around 15
years. SGI seems to be nicely focused on just turning out reasonable
systems without the b.s. First time in years I haven't had a vendor
telling me that, well, *I'm* not like their "real" customers (ya know,
if you do db's the vendor is more interested in CAD/CAM, if you do
science they wish you were a db site, etc)...this is refreshing.
AIX? You couldn't pay me enough...
--
-Barry Shein
Software Tool & Die | b...@world.std.com | uunet!world!bzs
Purveyors to the Trade | Voice: 617-739-0202 | Login: 617-739-WRLD
Casper H.S. Dik
>I used to be a big Sun/SunOS fan, owned a lot of Suns in my day (and a
>few Solbourne multis, Sparc, SunOS), but once they went to SVR4 it
>didn't seem to matter much compared to something like IRIX (also
>SVR4), and it was clear to me Sun was going thru pains changing over
>to SVR4 that I didn't want to share that with them. IRIX was about 2
>years ahead of them on SVR4 and, more importantly, SVR4/SMP. Maybe
>some day when they get it figured out I'll consider a big Sun again, I
>think Sun remains an interesting company, it's too bad they managed
>the switchover to SVR4 so poorly (IMHO, but follow the money.)
This strikes me as odd. SGI was mainly a SVR3, not SVR4 system It was
definitely not SVR4/SMP. SGI is SMP as has a more experience with that
than Sun. But SGIs MP servers have much the same problems as Suns.
Starting with IRIX 5.x SGI is moving toward SVR4 but isn't still their yet,
Solaris 2.x has more of SVR4 than SGI. (And I especially don't like the
way SGI dumped all of BSD in libc, which makes writing portable programs
on SGIs extremely hard, not to mention the fact that it each version
of ld(1) I used under IRIX 4.x was extremely broken to some extent)
I agree with the statement that Sun didn't manage the tranistion well.
It took so long, it seemed they had to release an unfinished product.
Solaris 2.x has gone a long way since and 2.3 has proven to be
very reliable for us.
Casper
Butch Deal
|> In article <3i05fo$l...@ra.nrl.navy.mil>,
|> Butch Deal <de...@ait.nrl.navy.mil> wrote:
|> > Suns with Solaris 2.x are about the easyest to maintain, configure, install,
|> > upgrade, etc, etc. and are about the most reliable.
|>
|> but the software is delightfully free of unexpected gotchas other than the
|> licensing ones. SunOS (Solaris 1.x) was reasonably good, if a little old.
|> Solaris 2.x has proven to be a nimgthmare, with no end of porting problems
|> and gratuitously different system administration interfaces. We have been a
|> System V shop for some time, with the SunOS stuff off the edge, and I'm pretty
|> familiar with BSD. Solaris isn't either. OSF/1 is pretty straight BSD with
|> a little System V admin stuff on the edge.
just enough to make it a pain in the...
this just makes it neither. You can't sit on the BSD/SYSV fence
forever.
It seems that all the people that do well with BSD just have a reall hard time
with solaris 2.x. Solaris 2.x is one of the closest SYSV OS's. I have
managed networks that were all BSD but I started out on SYSV and my personal
machines are SYSV, except one lowly sun3/60. Course I lone the 3/60 to a
friend and keep the SYSV machines for myself. :)
Dirk Grunwald
I think a lot of this is based on familiarity & what you've used in
the past.
On monday, I took delivery of three DEC multiprocessors (2100-500's,
aka Sables). By monday afternoon, I got one 2-CPU machine up, running
the AMD automounter, using BIND, doing SENDMAIL, exporting and
importing NFS files, using passwords and the like. This took about 1
hour, which was mainly involved in installing C++, Ada, DCE,
ObjectBroker (CORBA), etc etc from CDrom.
On tuesday, I cloned the next system & took delivery of 2 Alphastation
4/233's and started getting them set up.
By thursday, not only had I done all my other jobs (teach, serve on
mindless committees), but we had already sucked down massive
simulation cycles on the alpha cluster, including the one that
dual-boots and an OSF/1 and W/NT box.
I found V3.0B easy to deal with. I used to use (& manage) Sun machines
many eons ago, and I think the DEC stuff is pretty straight-forward.
I admit, I didn't do any fancy stuff like set up the RAID file system
(I did use the DEC log-structured file system), but really, to argue
that OSF/1 "isn't unix" or isn't robust is pretty silly. It's more
UNIX-like than anything I've used in 15 years of UNIX-dom, but some
parts are better.
Butch Deal
|> Butch Deal (de...@enterprise.ait.nrl.navy.mil) wrote:
|> : Suns with Solaris 2.x are about the easyest to maintain, configure, install,
|> : upgrade, etc, etc. and are about the most reliable.
|>
|> 1000 came online (it arrived in August and they spent the time inbetween
|> updating/customizing it). It has 1/4 GB of real memory, 2 CPUs, and a
|> "storage array". It ONLY handles our email (PINE) users (ie. no compilers,
|> stats, database, etc. software - JUST email).
|>
How many disks in the Sun RAID?
is it NFS serving for other systems?
is it a router?
does it serve DNS, NIS+, NIS, WWW, ftp?
|> The machine will grind to almost a halt at between 60-80 users - but it can
|> run okay with 120 users - but only for a while. It was running Solaris 2.3
|> in January, but they upgraded to 2.4 in an attempt at "fixing up" the problem
|> we are having. 2.4 did not fix up anything.
If the machine is really only used for email, I would check to make sure there
are no mud's running on it, etc.
check what your network load. you may be overloading it and not the machine.
you say 60 to 80 users and they only run pine.
clarknet (a local provider in the area) has a SS20/612 that can handle that
kind of load with no problem. It is still running solaris 2.3. They plan
to upgrade it to 2.4 and 712 cpu's soon though. There is almost never less
than 50 users on their machine and usually over 100 users. The users on
that SS20 do not just run mail either. Do you have patch 102119-01 installed?
what other patches do you have installed?
|>
|> So, and again, this is not a flame - if people with 5+ years of experience
|> with SUN systems/technology cannot get a (about the same as a model 590)
|> SPARC 100 working well after 6 months of trying, just how "simple to take
|> care of" can those systems be?
not a flame here either but just because someone has 5 years of experience
does not mean they know all that much. I have turned down many applicants with
more years of experience because they were not up to speed with modern tech.
|>
|> Any and all suggestions welcomed! :-)
|>
|> Ron,,,
|> Who only has three choices - if I can't use VM, then I'll settle for AIX! :-)
Whitney de Vries
file systems on other OS's is like taking a walk back into the
past when unix used to be a Ken Thompson's OS for games.
-- Whitney
ae...@cerfnet.com
>Well there must be some crazies out there. You can buy one now from Aspen
>Systems and rumor has it that Nekotech is finishing up the firmware that
>will let their new systems run it too (possibly also available as an
It's all moot anyway, 5 years from now *all* UNIX machines will be running
the Intel-HP PA-RISC-P8 chip, just because of economies of scale... :-)
Tony Burzio
AETC
San Diego, CA
Dave Olson
| (And I especially don't like the
| way SGI dumped all of BSD in libc, which makes writing portable programs
| on SGIs extremely hard, not to mention the fact that it each version
| of ld(1) I used under IRIX 4.x was extremely broken to some extent)
We can't win on this one. When we had a seperate libsun and libbsd, we had
an unending stream of complaints about this 'non-standard' arrangement
(literally thousands of complaints and questions on the net and to support).
Oh well.
--
The most beautiful things in the world are | Dave Olson
those from which all excess weight has been | Silicon Graphics
removed. -Henry Ford | ol...@sgi.com
ae...@cerfnet.com
>2) The lp services have a number of problems. The
>/var/spool/lp/system/pstatus file is not correctly formatted. By default,
>the banner page is turned on _AND_ turning it off in admintool doesn't
>turn it off _AND_ Sun tech support couldn't even tell me how to turn it
>off.
Sun has Tech Support? Could'a fooled me!!! I've called them numerous times to
fix Solaris problems, and I never get a call back. I think Sun Support is just an
answering machine :-( (Yes, I have a support contract, whatever that means to
Sun. I guess it pays for the box-fulls of patches for Solaris I get every other week... :-)
By the by, HP-UX is the best :-)
Tor Lillqvist
until now, this new brand of machine we just got sucks."
--
Tor Lillqvist,
working, but not speaking, for the Technical Research Centre of Finland,
Information Technology
Frans van Hoesel
: I think a lot of this is based on familiarity & what you've used in
: the past.
: On monday, I took delivery of three DEC multiprocessors (2100-500's,
: aka Sables). By monday afternoon, I got one 2-CPU machine up, running
: the AMD automounter, using BIND, doing SENDMAIL, exporting and
: importing NFS files, using passwords and the like. This took about 1
: hour, which was mainly involved in installing C++, Ada, DCE,
: ObjectBroker (CORBA), etc etc from CDrom.
: On tuesday, I cloned the next system & took delivery of 2 Alphastation
: 4/233's and started getting them set up.
: By thursday, not only had I done all my other jobs (teach, serve on
: mindless committees), but we had already sucked down massive
: simulation cycles on the alpha cluster, including the one that
: dual-boots and an OSF/1 and W/NT box.
This easely to improve on when installing SGI.
1) the system comes pre installed
2) when booting the first time (after only 5 minutes of unpacking)
it asks you for a machine from wich you want to use the configuration
ans - asuming you have one configured - it works. in ten minutes
3) installing the other stuff (C++ fortran, nfs etc) takes another
hour (because the cd ain't any faster; not because the work involved
which is nearly nothing)
4) starting automount, exporting disk etc is just a matter of a few
click with your mouse in the graphical admin programs.
Using a remote printer is as simple as dragging it from the find-tool
(which locates resources on other machines) onto your desktop. The same
method can be used with disk or remote tape drives.
So infact your are ready in ten minutes with the extra stuff in 2 hours.
you know what, I'm biased because I love their machines.
-frans
Frans van Hoesel //// hoe...@chem.rug.nl
pme...@unix.atk.com
In article <3i05fo$l...@ra.nrl.navy.mil>, <de...@enterprise.ait.nrl.navy.mil>
writes:
> In article <3hvfrk$c...@osiris.wu-wien.ac.at>, Rupert Thurner
<rthu...@x11srv.edvz.uni-klu.ac.at> writes:
> |> wa...@suni202.larc.nasa.gov (C Wayne Huling) wrote:
> |> what do admins of other systems think of this?
> |>
> |> rupert
SGIs bite!
My staff and I manage a collection of about 40 SGIs, 40+ HP700s, 10+ HP800s,
and about 90 Suns. I personally spend about 5 hrs a week on the Suns, 2 other
admins work on the HPs, and myself and another admin work on the SGIs. The
man-hours devoted to HPs are a little skewed because on the 800s and about
10 of the 700s we run "production" databases that we seem to be constantly
moving around --both the database and the hardware. The other 30 or so 700s
are in use for engineering applications and don't need much attention.
The SGI and Sun hours are also a little skewed because a bunch of them are
in "closed" secure areas --so no network, auditd running, we have to go in
there to work on them, etc.
Anyway, we spend a disproportionate amount of time on SGIs just keeping them
running. We have been trying to migrate to 5.2 and then 5.3 for a few months
now without much success. The objectserver has a memory leak (real bad in 5.2,
not so bad in 5.3) that requires us to reboot the boxes on a very regular
basis (once a day at 5.2, once or twice a week at 5.3). SGI tells us that we
are unique with the problem at 5.3 and it has something to do with our NIS
environment. I have a hard time believing that our environment is unique.
All our boxes (Sun, HP, SGI) are in the same NIS domain and we make heavy
use of the automounter, so you can log into anything and you only have one
homedir. We have a few Indys which only run version 5, and 4.0.5whatever was
not really intended to be run on Indigo 2s, so we really need to do this
migration. We have replaced about 15 monitors, 4 motherboards, and a power
supply on the 18 Indigo 2s we have. We also spend about $80,000 annually on
SGI software support. Yikes. I am happy with all of our older SGI hardware
running 4.0.5x, they run like champs with little or no attention needed, its
just the new stuff that sucks.
My gripes with HP mainly have to do with their high cost and trouble getting
third party stuff to work on their hardware --although they have been getting
much more "open" lately. I also tend to complain about the difficulty getting
common freeware tools to compile --although it is getting much easier and
there is a great archive site at ftp.cae.wisc.edu for already ported HP stuff.
I don't have too many complaints about Sun, but that probably is because we
haven't done the Solaris thing yet --we haven't needed to. But because I
haven't upgraded yet, I will probably benefit from all the pain that others
have gone through. I have been running Solaris 2.x for about a year on my
workstation, but I'm the only one who has it. I should be well trained by
the time we do migrate, and I don't forsee a huge problem.
SGI does a great job running ProEngineer and some of our mechanical analysis
tools, but the admin time on them is big. Sun CPU performance stinks, but
I am hoping they do something about that before our price/performance manager
types overrun us with HP and SGI purchases ( 3 news Suns, 20+ SGIs, 20+ HPs
in the last three years). I would have to give HP the thumbs up for overall
usability. They kick butt on the apps we run, and their 9.0[3,5] is pretty
easy to take care of. Lets hope they don't screw it up with version 10.
Barry Shein
From: cas...@fwi.uva.nl (Casper H.S. Dik)
>b...@world.std.com (Barry Shein) writes:
>
>>I used to be a big Sun/SunOS fan, owned a lot of Suns in my day (and a
>>few Solbourne multis, Sparc, SunOS), but once they went to SVR4 it
>>didn't seem to matter much compared to something like IRIX (also
>>SVR4), and it was clear to me Sun was going thru pains changing over
>>to SVR4 that I didn't want to share that with them. IRIX was about 2
>>years ahead of them on SVR4 and, more importantly, SVR4/SMP. Maybe
>>some day when they get it figured out I'll consider a big Sun again, I
>>think Sun remains an interesting company, it's too bad they managed
>>the switchover to SVR4 so poorly (IMHO, but follow the money.)
>
>This strikes me as odd. SGI was mainly a SVR3, not SVR4 system It was
>definitely not SVR4/SMP. SGI is SMP as has a more experience with that
>than Sun. But SGIs MP servers have much the same problems as Suns.
Yer showing yer age, how long has it been since 5.x started shipping?
A few years?
At any rate, the new IRIX 5.3 seems to have reached a new level of
maturity in SMP, parallelized networking internals, etc. They still
need to do some work on their pty performance (probably actually
related to streams) but this is fine with 200+ people online. I hear
Sun's SS2000 systems (comparable in market niche to my Challenge XL)
can't do anything like that tho Solaris 2.4 may have helped, adding
CPUs doesn't help (ie, money doesn't help, that's frustrating!) There
are also a lot of niceties like 8GB partitions (and an add-on from
SGI, XFS, that cranks that up to some unthinkable number as well as
adding some other features, ask them.)
I also like being able to go to 36 CPUs on the Challenge if and when
the day comes, vs 20 on the SS2000, tho I'll guess that likely when
that day comes Sun has figured out a way to double the CPUs so they'll
be comparable. In all honesty a max of 20 would serve us right now and
for the next year or so, but hey I like big.
>Starting with IRIX 5.x SGI is moving toward SVR4 but isn't still their yet,
>Solaris 2.x has more of SVR4 than SGI.
One man's bug is another man's feature :-)
>(And I especially don't like the
>way SGI dumped all of BSD in libc, which makes writing portable programs
>on SGIs extremely hard, not to mention the fact that it each version
>of ld(1) I used under IRIX 4.x was extremely broken to some extent)
Well, forget 4.x, we may as well be talking about SunOS 4.x bugs if
that's the measure.
>Solaris 2.x has gone a long way since and 2.3 has proven to be
>very reliable for us.
Are you running any large-ish multis? SS1000s or SS2000s?
Ron Wigmore
So, I get to type it all in again! :-(
Butch Deal (de...@enterprise.ait.nrl.navy.mil) wrote:
: In article <3i0rl2$14...@hermes.acs.ryerson.ca>, rwig...@acs.ryerson.ca (Ron Wigmore) writes:
: |> Not a flame by any means, but at the start of January our brand new SPARC
: |> 1000 came online (it arrived in August and they spent the time inbetween
: |> updating/customizing it). It has 1/4 GB of real memory, 2 CPUs, and a
: |> "storage array". It ONLY handles our email (PINE) users (ie. no compilers,
: |> stats, database, etc. software - JUST email).
: How many disks in the Sun RAID?
"iostat -x 30" says 21 (I had thought there were 24?) Only two are running
with busy/svc_t level above 30%, 50 ms. That'll turn into a problem once we
can get > 200+ people on the box, but right now, it is not *the* problem.
Some disks are not being used yet, but it is a new box so ...
: is it NFS serving for other systems?
Yes, about 5 GB (each) from 2 RS6000 model 580s and it has about 8 GB NFS
mounted on each 580. nfs_ninode has been set to 20000. The 580's will have
100-150 users (each) when they too end up getting high load averages. Last
semester that was normal, since everyone could only login to the 2 580s. I
think it's safe to say that activity on one box can slow down another, but
not always. And, a 580 can end up with high load averages even when the
SC1000 is running fine, but it is more that the SC1000 gets busy, and then
a 580 gets busy.
"netstat -s" says there are a lot (10%+) fragmentation/reassemble/duplicate
packets and udp socket overflows. I'm going to have them reduce the rbuf
and wbuf sizes down to 2K. There were/are token ring "xmit/recv" type errors
being issued - they had them checked out, and they said everything was okay
with the SC100's token ring adapter. The network is okay - tuning on the
580s got rid of "congestion errors", but the SC1000 is still getting some.
"netstat -s" on the 580s gives "nice" numbers now.
The "dmesg" command shows:
$Feb 18 16:36
$tr0 at sbi0: SBus0 slot 1 0xc00000 and SBus0 slot 1 0x0 and SBus0 slot 1 0x400000 SBus level 4 sparc ipl 7
$tr0 is /io-unit@f,e0200000/sbi@0,0/tr@1,c00000
$ip_rput: DL_ERROR_ACK for 29, errno 1, unix 0
$ip: joining multicasts failed on tr0 - will use link layer
$broadcasts for multicast
I've been told those are "okay", but "error messages make me nervous", so...
: is it a router?
No - not a router
: does it serve DNS, NIS+, NIS, WWW, ftp?
Only FTP - either to/from other sites, or up/down from PCs.
Although, we do have 18K users in our password file. I think that might be
*a* factor since logins/rsh seem to take a while. AIX lets us use "hashed"
passwd files - I'm told Solaris does not. Last semester they used to use NIS
(580s only), but stopped due to security and network load reasons.
Would running with NIS+ - only on the SC1000, the 580's would not be NIS
slaves - gain us any advantages from using NIS's "hashed passwd" files?
NB: NIS is *really* nice, but the performance/response times seems to be
inversely proportional to the number of logged in users.
: |> The machine will grind to almost a halt at between 60-80 users - but it can
: |> run okay with 120 users - but only for a while. It was running Solaris 2.3
: |> in January, but they upgraded to 2.4 in an attempt at "fixing up" the problem
: |> we are having. 2.4 did not fix up anything.
: If the machine is really only used for email, I would check to make sure there
: are no mud's running on it, etc.
Only email and (oops!) news (tin). A (nologins allowed) 350 handles the news
feed and DNS. News can only be read from the SC1000. The SC1000 receives
all incoming mail, but it and the 580s can send out mail.
"w" says there are no "heavy CPU users" on the box - during the high load
periods. "top" says it isn't user generated load, either.
: check what your network load. you may be overloading it and not the machine.
The network (token ring only - no ethernet) traffic is about 150 p/s, with
peaks of maybe up to 500. The 580's have been tuned to handle 1500 p/s by
"pre-allocating" the needed "network" resources. "tr_nbufs" has been set
to the maximum(?) of 400.
Are there no other tuneables for network (non-NFS) stuff on Solaris? AIX
lets us set extra stuff. Our SC1000 can die with 60 users, but run fine
with 120. I'd say it's on the network side (within the SC1000) of things
and when "a condition/event" occurs, that's when the SC1000 goes crazy in
its efforts to keep things running. I formed that opinion when we were at
2.3 (CPU bound, "mutex stalls" were the norm), but now we just keep getting
load averages of 25-100 when we have 60 people just reading email! :-(
: you say 60 to 80 users and they only run pine.
: clarknet (a local provider in the area) has a SS20/612 that can handle that
: kind of load with no problem. It is still running solaris 2.3. They plan
: to upgrade it to 2.4 and 712 cpu's soon though. There is almost never less
: than 50 users on their machine and usually over 100 users. The users on
: that SS20 do not just run mail either.
Let 'em know the stats are different! "mpstat" was showing "mutex stalls"
when we were running 2.3 - also, the number of minor faults were much higher
then with 2.4! I'm not too certain what other numbers are different - I've
only been working with Solaris since 1995. :-)
: Do you have patch 102119-01 installed?
I'll ask on Monday.
: what other patches do you have installed?
Ditto. :-)
(diplomatic comments with which I agree deleted! :-))
Ron,,,
I've read Adrian Cockcroft's book - a few times - I hope I'm not embarassing
either of us! :-)
Vic Walker
>> I have admined every major workstation (Sun3, Sun4, SGI, DEC, HP, IBM,
>> DEC Alpha, DEC Station, Linux, SCO Unix) and let me tell you DEC ranks
>> real low on my list...
And Vic replies:
Wayne, have you worked with NeXTStep (I found this note in a NeXTStep
newsgroup), and could you give me a comparison with linux? I'm trying to
set up a NeXTStep-based UNIX network for a high school district. Trying
to get NeXTStep to do ANYTHING I want has been like pulling teeth. I have
set up Linux on my own computer, and have found installation, maintenance,
and just plain working on the computer to be much easier with Linux than
NeXTStep. Have you (or any of you other gentle readers) had similar
experiences?
Our main reason for using NeXTStep is because NeXTStep is commercial, and
so ought to be better supported than Linux (not sure that's true, though,)
and because it supposedly has built-in support for database management.
After working with NeXTStep for several months now, I'm beginning to think
we made a mistake.
Your opinions, folks?
Thanks much. Please respond via E-Mail, as well as in this newsgroup,
because I don't get over this way that often.
Vic Walker
vwa...@netcom.com
fishbowl
Rupert Thurner <rthu...@x11srv.edvz.uni-klu.ac.at> wrote:
>wa...@suni202.larc.nasa.gov (C Wayne Huling) wrote:
>>
>>
>> I haven't seen much mention of DEC here yet. Which brings me to a point,
>> why are so many people religiously attached to Sun's?
Suns have widespread support, for applications that (gasp) aren't available
for linux. If you can put the gnutils, a decent shell, and proper X11R6
instead of openwin, it's a nice stable platform with great horsepower,
awesome network performance, and high-performance video hardware (not to
mention the big flat screen :-)
>> search for a new workstation, they rank at the bottom of my list for
>> performance and OS reliability.
I wish I could get SUNos, Slowlaris is sucking here, in a big way. Sun
wants more than my budget for an upgrade. I've moved my archive / w3 server
to a linux 486dX100 with an adaptec which appears to outperform the sparc-5
significantly.
I would like to run a filesystem benchmark, but I can definitely say that the
local bus adaptec and a 1g Seagate out dos the sparc significantly.
I would like to run a filesystem benchmark.
>>I have not yet gotten to play with a Dec
Ditto, but one of our consultants does. The thing about RISC machines is
that to do the same work, they require that you have much more memory than
on x86 machines. I know that among the first things I want to know before
buying a workstation would be: Does it use regular, available on the street
at street prices, 72-pin SIMM's, or does it use some proprietary chip? On
a RISC machine, you are going to need RAM; and thanks to the PC market, some
RAM is a lot cheaper than others. I also like the idea of having disk drives
interchangeable with the rest of my SCSI boxes. Consumer-market SCSI drives
are fast, reliable, and relatively very inexpensive.
>> I have admined every major workstation (Sun3, Sun4, SGI, DEC, HP, IBM,
>> DEC Alpha, DEC Station, Linux, SCO Unix) and let me tell you DEC ranks
>> real low on my list. The recent release of OSF1 V3.0 is a licencing
>> nightmare. Only two users are allowed to login at once (root does not
>> count) under the default licence. A licence has to be purchased for
>> each additional user. These licences are yearly, so good paperwork
>> has to be kept on the machine. Then, the little quirks of keeping
>> them running.... I have a DEC alpha loaded with 3.5 internal drives,
>> and it has a heat problem. This was a suprise to the DEC Engineers,
>> and they only offered advice was move it to a colder room or get some
>> of the drives out. We also run some pretty intensive programs on the
>> DEC, and a single STAGS (Statistical Analysis Program) can bring the
>> machine to its knees, with nothing else running (above system
>> software). The DEC support people have alot to learn about customer
>> support!! The ONLY DEC persons who seemed to want to help when I
>> called was a Licencing person (Chris Kelley) who helped me through my
>> nightmare of upgrading licences.
>>
>> If you want a really good price/performance ratio, I would
>> suggest a Pentium based machine. A P5-100 pulls in SPEC numbers better
>> than the entire Sparcstation 5 line
My $0.02 on this, I think the 486/100 will also. In my application it
definitely does. Overall performance is greater than a P5 box we have. The
bottleneck is the memory bandwidth and the disk controller. PCI disk transfer
rate is faster than anything I've seen, I must admit.
>> I would appreciate hearing anymore that anyone has to say on this subject.
I hope you don't mind my biased ravings. I'm having a find time
to X11r6 finally running on my home machine! And on the net full time!
>> A lot of people like Suns because the Administration of them
>> is rather easy.
That almost made me cry! Don't let my boss see that?! In a relative way, if
you can say ANY \u\n\i\x administration deserves the moniker of 'easy'
As far as I'm concerned, it is always a challenge. The learning curve never
stops...
SGIs, a little more difficult (IMHO), but I have
Now I am gonna cry! We're getting one or maybe two SGI's in, and I will
have to deal with them!
>> one of the Developing programmers helping me with my problem.
They are rightfully proud of their achievements.
>> I want machines that have proven
>> reliability. For me, Suns and SGIs are at the top of the list....
My linux servers and my sparc's are competing for uptime records. Most of
the reliabiliy problems in my shop result from human error. I find the
linux systems to be easy to "see" and work with; system V, very lofty and
mysterious. I have to be superstitious because I'm always logging in as su. on
a gateway/nameserver with 800 users.
In my opinion, linux b/w gnu software is a relatively easy system. I believe
it leads to fewer errors made by administrators. (Me)
If I could find a combination of ethernet card/driver/kernal pl that satisfied
me, I would put the pc against the workstation any day. As you see, my
applications do not require massive speed, but they do benefit from
'horsepower.' The nature of the motorola architechture is simply better
in many respects to (What should I call it? PC? intel? isa?)
Sincere, sleepy regards,
James McGill
1:124/90...@fidonet.org
Peter da Silva
Butch Deal <de...@ait.nrl.navy.mil> wrote:
> Suns with Solaris 2.x are about the easyest to maintain, configure, install,
> upgrade, etc, etc. and are about the most reliable.
Hmmm. Not my experience. I'm not really thrilled at DEC's system admin tools,
but the software is delightfully free of unexpected gotchas other than the
licensing ones. SunOS (Solaris 1.x) was reasonably good, if a little old.
Solaris 2.x has proven to be a nimgthmare, with no end of porting problems
and gratuitously different system administration interfaces. We have been a
System V shop for some time, with the SunOS stuff off the edge, and I'm pretty
familiar with BSD. Solaris isn't either. OSF/1 is pretty straight BSD with
a little System V admin stuff on the edge.
> DEC's don't
> really count in the comparison till they get a little closer to UNIX.
If Solaris is your definition of UNIX, I'd love to see UNIX get a little
closer to DEC.
--
Peter da Silva `-_-'
Network Management Technology Incorporated 'U`
1601 Industrial Blvd. Sugar Land, TX 77478 USA
+1 713 274 5180 "Hast du Heute schon deinen Wolf umarmt?"
Paul D. Robertson
Motorola currently makes a PREP line that runs AIX 4.x, as well as
(*ugh*) NT.
Bull either makes or re-badges RS/6K stuff too.
Paul.
: -art stine
: visix software
--
-----------------------------------------------------------------------------
Paul D. Robertson "My statements in this message are personal opinions
prob...@clark.net which may have no basis whatsoever in fact."
PSB#9280
Dirk Grunwald
In article <3i5odn$g...@rugch4.chem.rug.nl> hoe...@chem.rug.nl (Frans van Hoesel) writes:
FvH> So infact your are ready in ten minutes with the extra stuff in 2
FvH> hours. you know what, I'm biased because I love their machines.
--
Well, in fact, the systems came pre-installed with OSF/1 V3.0B. The
extra time was installing the layered products (i.e., things not
normally instlled, like the CORBA implementation), and installing
local things (e.g., we use the 'amd' automounter locally rather than
the 'automount' mounter).
So, in fact, the system "worked" in 10 minutes, but it didn't conform
to our local conventions & didn't have Corba, ada or C++ installed.
Will Morse
Steve Strange <str...@zk3.dec.com> wrote:
>In article <3i05fo$l...@ra.nrl.navy.mil> de...@enterprise.ait.nrl.navy.mil (Butch Deal) writes:
>
>> would have to rate then higher than HP's and AIX systems though. DEC's don't
>> really count in the comparison till they get a little closer to UNIX.
>
>
> Steve
>--
>Steve Strange
>Digital Equipment Corporation, Nashua, NH
>DCE DFS for DEC OSF/1 -- Development
>str...@zk3.dec.com
Is it true that Solaris 2 is really VMS under the covers? :) :) :)
Will
--
# Gravity, # Will Morse
# not just a good idea, # BHP Petroleum (Americas) Inc.
# it's the law. # Houston, Texas
# # wi...@starbase.neosoft.com
#
# These are my views and do not necessarly reflect the views of BHP !
Will Morse
David Mathog <mat...@seqvax.caltech.edu> wrote:
>In article <3i2hrv$a...@Starbase.NeoSoft.COM>, wi...@Starbase.NeoSoft.COM (Will Morse) writes...
>>and no one in their right mind would make an Alpha clone.
>
>Systems and rumor has it that Nekotech is finishing up the firmware that
>will let their new systems run it too (possibly also available as an
>or VMS, and if you've got the bucks for three different system disks and
>all of those OS's, you can switch between them on the same machine by just
>rebooting. Kind of handy for developers who work in all three OS's, they
>only need one workstation instead of three.
>
>Regards,
>
>David Mathog
>mat...@seqvax.bio.caltech.edu
>Manager, sequence analysis facility, biology division, Caltech
Well, at least IMHO, you're right, there are crazies out there.
To make it as a clone vendor you have to have some advantage (delivery,
configuration options, whatever) plus a price advantage of about 30%.
There has to be a big enough market in the first place for you to take
maybe 2-5% of that and still make money. You can save money by bending
your own sheet metal and making your own boards, but beyond that, the chips
and stuff cost pretty much the same. You might be able to save money
on labor, but the labor-saving advantages of insertion technology, robots,
etc. tend to accrue to the volume leader.
DEC is pretty much selling OpenVMS (aka EmptyVMS) systems to its installed
base. Maybe they are selling OSF/1 machines to some small crowd, but
regardless, there is no comparison to the market share of a Sun or
Intel-based leader. So the question is, what are Aspen Systems or
Nekotech going to get 2-5% of and how is that going to pay off better
than putting that effort into yet another Pentium system?
DEC spent a lot of time, money, and effort pissing off the Unix community
in the not too distant past. There are a lot of companies that aren't
going to let them back in very easily. They are going to have to make
sales on mission critical third party software, and I don't think they
are making the sales to these developers that they did in the old VMS
days.
On the other hand, who would have guessed tha Sun would fail to go for
the jugular by discontinuing their solid, robust, popular operating
system and going to a new system, now in its fifth beta release, just
in time for DEC to survive long enough to get back into the fight?
I believe that it is not enough to be the best bang for the buck. To
make it these days you have to be the lowest price in the minimum
performance range that will get your customer's job done. Sun is NOT
successful because they are good. They are successful because they
are _good_enough_ and cheap. This success lets clone makers in because
they can be _just_as_good_enough_ and even cheaper. DEC is, at least
at present, NOT good enough and certainly not cheap enough. This
gives clone makers no where to go.
As to whether Aspen and Tkotech are crazy, I guess that's for their
financial backers to decide.
Peter Shenkin
Dave Olson <ol...@anchor.engr.sgi.com> wrote:
>cas...@fwi.uva.nl (Casper H.S. Dik) writes:
>| (And I especially don't like the
>| way SGI dumped all of BSD in libc, which makes writing portable programs
>
>We can't win on this one. When we had a seperate libsun and libbsd, we had
>an unending stream of complaints about this 'non-standard' arrangement
>(literally thousands of complaints and questions on the net and to support).
If this is a battle that's worth winning, one way to go would be to
supply a "-svr4" load flag. This would load a library purged of bsd-isms.
This idea is very much in the spirit of the "-ansi" flag to "cc".
-P.
--
************************ The secret of life: *************************
*Peter S. Shenkin, Box 768 Havemeyer Hall, Chemistry, Columbia Univ.,*
* New York, NY 10027; she...@columbia.edu; (212) 854-5143 *
************* If you find a loose thread, don't pull it. *************
Ronald S. Woan
Stefan Monnier <mon...@di.epfl.ch> wrote:
>Well, AIX isn't the only unix with an advanced file system.
>It's actually getting common to offer a log-structured FS addtionnally
>to plain UFS. It's so much faster to reboot and to repartition !
IBM was the first of the big players to offer this (contributed this
to OSF) and in the latest rev, we also offer inline compression along
with some more tuning parameters.
--
+------All Views Expressed Are My Own And Not Necessarily Shared By IBM-----+
+ Ronald S. Woan (IBM VNET)WOAN AT AUSTIN, wo...@exeter.austin.ibm.com +
+ outside of IBM wo...@austin.ibm.com or Compuserve: 73530,2537 +
+ URL: http://cactus.org/~woan/ +
Martin Cracauer
>cas...@fwi.uva.nl (Casper H.S. Dik) writes:
>| (And I especially don't like the
>| way SGI dumped all of BSD in libc, which makes writing portable programs
>| on SGIs extremely hard, not to mention the fact that it each version
>| of ld(1) I used under IRIX 4.x was extremely broken to some extent)
>We can't win on this one. When we had a seperate libsun and libbsd, we had
>an unending stream of complaints about this 'non-standard' arrangement
>(literally thousands of complaints and questions on the net and to support).
This is definitivly a major difference between Sun and SGI. Solaris
enforces the user to use only the newest interfaces, SGI does not.
I had to commit major changes on my own applications when I first used
a solaris machine, but all required changes are moves to *better*
calls. Better especially means that unsave functions/systemcalls are
avoided and replaced by modern couterparts (example: functions where
string pointers are passed and no length argument is supplied are
thrown out of libc).
Although I got into some trouble because the Solaris port requires
significant more time than any other UNIX port I did, I think Sun's
way is right. I don't remember a required change that was not an
improvement for my code (former BSD-based).
Another example: I had trouble with the SGI's default malloc that is
build into libc. There is a `malloc (3X)' package in a seperate
library and that is obviously better. If I inteprete SGI's behaviour
right there is customer code that relies on some undocumented feature
of the old malloc, so SGI didn't exchange the old code. I think Sun
would have done the other way and force the customer to change its
code. Again, I feel better with Sun's way.
I can understand why people hate Sun, it is possible to miss deadlines
because of changes in Suns environment. Howver, I prefer to do the
right thing as soon as possible (I'm can type fast :-).
[followups narrowed. My good, I hope the original poster is already
toasted...]
To comment on the original question:
I think Suns are best when you have to run a large network and don't
have any special requirements for the single machine. They're cheap,
easier to administer when it comes to many machines at once and you
get all kind of software and interface cards for them (for example,
combi cards with ethernet and SCSI or more than one ethernet in one
slot are not availiable for SGI).
I think SGIs are good machines for individuals that run their own
machine and want/need some of SGI's extra features. Of course, doing
graphics is a strength of SGI. Be warned that SGI suport and options
like SCSI-adapters, compilers etc. are much more expensive than for
Sun.
Both Sun and SGI have less floating point power than SGI/DEC/IBM.
If you need floating point power, and you want central servers for FP
work, use a network of Suns and some large DEC Alpha machines for the
FP work. The DEC machines usually have the best price/performance
rating, but maybe something does wrong and you cannot use them anymore
(DEC collapses or unforseen, but neccecary software doesn't run). In
that case you can easily exchange a central server. Exchanging a large
network of workstations and force users to change the environment on
their local screen doesn't happen with this solution.
If you want the fastest stuff around everywhere and/or don't want to
drive a heterogenous network, use HP. As for SGI, these boxes are much
more expensive when it comes to extentions and software. Most obvious,
only the large machine are as expandable as Sun's low-end
boxes. Additionally, their OS is stable, but misses some newer
features and is strange in many ways. It can cause headaches if you
have applications that are not already ported to HP (freeware or you
own).
IBM's strength is with some special kinds of Floating-Point
applications. The RS/6000 (not the PowerPC-based systems) have small
caches and high memory bandwidth and the processor is capable of doing
multiplication and addition at once. If you have one specific
application, test it on a RS/6000. If it meets the architecture's
strength, the RS/6000 has probably the best price/performance rating
of all machines. I would avoid AIX when this is not the case.
--
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Private email Martin....@wavehh.hanse.de Fax +4940 522 8536. No NeXTMail!
No guarantee for anything. Anyway, this posting is probably produced by one
of my cats stepping on the keys. No, I don't have an infinite number of cats.
Steve Kappel
If a 20 processor SS2000 isn't big enough the Cray 6400 (I think that is the
model) goes to 64 SPARC processors on 4 XDbus instead of 2 running
essentially the same O/S.
--
______________________________________________________________________
Steve Kappel Work: ste...@apertus.com
Apertus Technologies, Inc.
7275 Flying Cloud Drive
Eden Prairie, MN 55344
Mark Brown
>Spec 1170 is now supposed to be the determining factor of "unix or not", but
>when I hear things like IBM's plans on getting OS/400 to pass 1170, it makes
>me wonder. [don't know anything about os/400, except that I wouldn't think it
>Unix]
Welllll, let's talk about this. UNIX is supposed to be an "open" system,
huh? Well, to me that means a published set of APIs and behaviors that
*anyone* can implement.
If OS/400 (AS/400) can meet the published requirements (and BTW,
"Spec1170" is about as SVID3 as you can get, more like SVID3++), and
apps written for UNIX can prot and run on it with little or no change
(the same state that different UNIX flavors are in today)....
...then what is the problem?
1170 covers the basic *commands*, so that shouldn't be it.
It covers the basic header files and dev tools....
Sys Admin is different? Big Deal. Most every non-UNIX-geek will tell you
that the current state of UNIX administration is summed up in one word
"incomprehensible". Let competition flourish, maybe someone will come up
with something better. How many of us here actually use and like the
SysV accounting disaster, that dates back to 5.0 and the Western
Electric Black Book?
File systems different underneath? Competition came up with JFS (and
others) -- let it continue unless you *want* to go back to "good old
UNIX" hope-fsck-gets-my-files-back.
Folks, that fact the systems like the AS/400 are moving towards UNIX
APIs is a *win*win*win*.
cheers,
mark
--
Mark Brown | AIX Architecture | Civil Liberty
(512) 838-3926 | 11400 Burnet Rd M/S 9582 | Through
T/L 678-3926 | Austin, TX 78758 | Complex Mathematics!
mbr...@austin.ibm.com -or- MBROWN at AUSVM6 |
Michael Sternberg
>It seems that all the people that do well with BSD just have a
>reall hard time with solaris 2.x.
I back that up definitely. Started with Ultrix (BSD), though ;-) We
have now Alphas with OSF/1 in the lab and I find them rather easy to
administer and doing work with. Apart from the licencing trouble
already mentioned here.
We got a Solaris machine as a front end to an analysis system here.
The printed doc kept strict silence on sys admin stuff like
networking or hardware access. Instead, it explains what a keyboard
is and what "cp" and "mv" do. The system related man pages appear
either terse or poetic to me. I was called to repartition the disk
because it contained no separate /var with its 16Mb root pretty full.
(Ok - this is not Solaris's fault).
Took some time to find the proper command:
fmthard - populate VTOC on hard disks
This command cannot even handle the output of its counterpart
prtvtoc(1m).
Greetings, Michael
--
"Who disturrrbs me at this time?" << Zaphod Beeblebrox IV >>
Michael Sternberg http://www.tu-chemnitz.de/~mst Theoretical Physics III
ster...@physik.tu-chemnitz.de LabPhone: +49(0)371 531 3148
Martin Cracauer
>Butch Deal (de...@enterprise.ait.nrl.navy.mil) wrote:
>: Suns with Solaris 2.x are about the easyest to maintain, configure, install,
>: upgrade, etc, etc. and are about the most reliable.
>Not a flame by any means, but at the start of January our brand new SPARC
>1000 came online (it arrived in August and they spent the time inbetween
>updating/customizing it). It has 1/4 GB of real memory, 2 CPUs, and a
>"storage array". It ONLY handles our email (PINE) users (ie. no compilers,
>stats, database, etc. software - JUST email).
>The machine will grind to almost a halt at between 60-80 users - but it can
>run okay with 120 users - but only for a while. It was running Solaris 2.3
>in January, but they upgraded to 2.4 in an attempt at "fixing up" the problem
>we are having. 2.4 did not fix up anything.
>So, and again, this is not a flame - if people with 5+ years of experience
>with SUN systems/technology cannot get a (about the same as a model 590)
>SPARC 100 working well after 6 months of trying, just how "simple to take
>care of" can those systems be?
>Any and all suggestions welcomed! :-)
I've seen many reports that Sun's big servers (SS 1000 etc.) don't
work good, especially that they're unstable with OS releases that are
OK on `normal' machines. I don't use big Sun machines and the
workstations I run definitvly don't have the problems that are
reported for the servers.
The question is: Are the high-end-servers of SGI (or HP, DEC) really
better or are there fewer bug reports because they're not used as
often?
I cannot resist to beleive that *all* such MP-high-end Servers have
problems and that the high-volume workstations run better for all
vendors.
I'd like to see success reports from SGI's challenge (or similar
machines from other vendors) here.
BTW, Sun's performance is worse than HP/DEC/IBM especially for
flaoting point. But a SPARC clone usually beats other workstations in
price/performance easily when you need mostly integer performance.
Martin
Bob Rein
>
> >We are currently looking at either at Sun or HP to run an Oracle database
> >application. Any comments on the SMP capabilities of Sun?
>
> I'm not running an SMP Sun right now. I think the best advice would be
> to find a few good reference sites to speak with, Sun or Oracle (or
> perhaps the net) should be able to point you at some.
I have received a little input on the net and it has been somewhat
favorable for Sun servers, but it appears users have not been too happy
with earlier versions of Solaris. It looks like version 2.4 is more
stable, but I believe they haven't ported the SAP manufacturing package
to 2.4 yet and are using Oracle 7.0.16.?
>
> My understanding is that that Sun considers their SMP systems as a
> platform for Oracle a high priority, so you're barking up the right
> tree. I wouldn't exclude SGI's Challenge systems in your review
> (unless there's some other reason.)
>
The SGI box looks pretty good, but I don't believe the SAP guys have
ported to that box or that SGI has not provided a competency center to
work with there product.
Butch Deal
|> Dirk Grunwald (grun...@foobar.cs.colorado.edu) wrote:
|>
|> : I think a lot of this is based on familiarity & what you've used in
|> : the past.
|>
|> : On monday, I took delivery of three DEC multiprocessors (2100-500's,
|> : aka Sables). By monday afternoon, I got one 2-CPU machine up, running
|> : the AMD automounter, using BIND, doing SENDMAIL, exporting and
|> : importing NFS files, using passwords and the like. This took about 1
|> : hour, which was mainly involved in installing C++, Ada, DCE,
|> : ObjectBroker (CORBA), etc etc from CDrom.
|>
|> : On tuesday, I cloned the next system & took delivery of 2 Alphastation
|> : 4/233's and started getting them set up.
|>
|> : By thursday, not only had I done all my other jobs (teach, serve on
|> : mindless committees), but we had already sucked down massive
|> : simulation cycles on the alpha cluster, including the one that
|> : dual-boots and an OSF/1 and W/NT box.
|>
|> This easely to improve on when installing SGI.
|> 1) the system comes pre installed
yeah with everything in one large partition!
|> 2) when booting the first time (after only 5 minutes of unpacking)
|> it asks you for a machine from wich you want to use the configuration
|> ans - asuming you have one configured - it works. in ten minutes
This is only for your personal window setup.
|> 3) installing the other stuff (C++ fortran, nfs etc) takes another
|> hour (because the cd ain't any faster; not because the work involved
|> which is nearly nothing.
make sure you install them in the right order though.
|> 4) starting automount, exporting disk etc is just a matter of a few
|> click with your mouse in the graphical admin programs.
|> Using a remote printer is as simple as dragging it from the find-tool
|> (which locates resources on other machines) onto your desktop. The same
|> method can be used with disk or remote tape drives.
it only locates recources on other SGI systems. The sgi solution to remote
printing is an rsh for the lp system.
|>
|> So infact your are ready in ten minutes with the extra stuff in 2 hours.
on a sun, unpack it, hook up the cables, plug it in, put the MAC address
in the ethers map and run add_install_client on the server, then turn on the
new sun. This is for a dataless configuration which most sun admins use
(saves alot of disk space and admin head aches). Diskless is similar. If you
whant the machine to be diskfull it is the same as dataless. Get the patch
scripts from Casper Dik, and you can add all the patches to the dataless and
diskfull clients as they install. In fact I just installed solaris 2.4 on 26
dataless Suns. Took about an hour. Good deal of that spent walking from room
to room to turn on the machines. Thats not just the base OS thats full
install, all the latest patches (downloaded this morning from sun), all the
sun add on software (compilers, SDK, WABI, etc) and all the third party
software.
|>
|> you know what, I'm biased because I love their machines.
|>
--
Butch Deal
Michael Sternberg <st...@physik.tu-chemnitz.de> wrote:
>We got a Solaris machine as a front end to an analysis system here.
>The printed doc kept strict silence on sys admin stuff like
>networking or hardware access. Instead, it explains what a keyboard
>is and what "cp" and "mv" do. The system related man pages appear
what printed doc's do you have? try the answerbook.
>either terse or poetic to me. I was called to repartition the disk
>because it contained no separate /var with its 16Mb root pretty full.
>(Ok - this is not Solaris's fault).
>Took some time to find the proper command:
>
> fmthard - populate VTOC on hard disks
>
>This command cannot even handle the output of its counterpart
>prtvtoc(1m).
try format. It works pretty much like the Sunos format program.
Barry Shein
From: crac...@wavehh.hanse.de (Martin Cracauer)
>The question is: Are the high-end-servers of SGI (or HP, DEC) really
>better or are there fewer bug reports because they're not used as
>often?
Feel free to telnet to world.std.com and login as new (no password)
and create an account. If you use it less than one hour we won't even
charge you anything.
World.std.com, the oldest public access internet system, is an SGI
Challenge XL with eight 200MHZ R4400 CPUs, 512MB, etc, running IRIX
5.3. The news server is an SGI Challenge L with four 100MHZ R4400,
256MB. They're attached across a FDDI. We currently have about 10,000
customers.
Seeing is believing.
>I'd like to see success reports from SGI's challenge (or similar
>machines from other vendors) here.
You got it.
Rupert Thurner
1. are "standard" machines (pc's) capable to keep track with
proprietary risc machines (servers - workstations).
2. is it worthful to bother about configuring unix if
winnt sets up without doing anything "manually" (nearly ;) )?
3. is the difference in ease of administration (if this is true)
and efficiency worth the price difference ?
4. whats about software for the different plattforms
(ease of use - stable - price - public domain)
Martin Knoblauch
(Butch Deal) writes:
|> In article <3i5odn$g...@rugch4.chem.rug.nl>, hoe...@chem.rug.nl (Frans
|> van Hoesel) writes:
|> |> Dirk Grunwald (grun...@foobar.cs.colorado.edu) wrote:
|> |>
|> |> : I think a lot of this is based on familiarity & what you've used
|> in
|> |> : the past.
|> |>
|> |>
|> |> 1) the system comes pre installed
|>
|> yeah with everything in one large partition!
|>
This is the second time that I see this kind of negative
remark on this particular topic in this thread. What is the
exact problem with the one partition layout? Definitely beats
a small root partition that fills up every second day :-)
Martin
--
+---------------------------------+-----------------------------------+
|Martin Knoblauch | Silicon Graphics GmbH |
|Application Center | Am Hochacker 3 - Technopark |
|Silicon Graphics Computer Systems| D-85630 Grasbrunn-Neukeferloh, FRG|
| | Phone: (+int) 89 46108-179 or -0 |
| | Fax: (+int) 89 46108-190 (-222) |
+---------------------------------+-----------------------------------+
|Network: <kn...@munich.sgi.com> | V-Mail: 5-8935 | M/S: IMU-315 |
+---------------------------------------------------------------------+
Eric Werme
>Dirk Grunwald (grun...@foobar.cs.colorado.edu) wrote:
: By monday afternoon, I got one 2-CPU machine up, ... This took about 1
: hour, which was mainly involved in installing C++, Ada, DCE,
: ObjectBroker (CORBA), etc etc from CDrom.
>This easely to improve on when installing SGI.
>3) installing the other stuff (C++ fortran, nfs etc) takes another
> hour (because the cd ain't any faster; not because the work involved
^^^^
> which is nearly nothing)
>So infact your are ready in ten minutes with the extra stuff in 2 hours.
^^^ ^^^^^^^
>you know what, I'm biased because I love their machines.
You must love some interesting clocks, too.
--
Eric (Ric) Werme | we...@zk3.dec.com
Digital Equipment Corp. | This space intentionally left blank.
Butch Deal
|> In article <3iarfd$c...@ra.nrl.navy.mil>, de...@enterprise.ait.nrl.navy.mil
|> (Butch Deal) writes:
|> |> In article <3i5odn$g...@rugch4.chem.rug.nl>, hoe...@chem.rug.nl (Frans
|> |> van Hoesel) writes:
|> |> |> Dirk Grunwald (grun...@foobar.cs.colorado.edu) wrote:
|> |> |>
|> |> |> : I think a lot of this is based on familiarity & what you've used
|> |> in
|> |> |> : the past.
|> |> |>
|> |> |>
|> |> |> This easely to improve on when installing SGI.
|> |> |> 1) the system comes pre installed
|> |>
|> |> yeah with everything in one large partition!
|> |>
|>
|> This is the second time that I see this kind of negative
|> remark on this particular topic in this thread. What is the
|> exact problem with the one partition layout? Definitely beats
|> a small root partition that fills up every second day :-)
|>
There are a few places all over that anyone can write to. When you have
one partition, what you really have is a large root partition with several
places anyone can write to. Home directories are in the one partition as well.
pretty easy till fill up the large partition with your user partitions there
as well as a few temp dirs.
Paul 'Shag' Walmsley
masochistic ..
Butch Deal (de...@enterprise.ait.nrl.navy.mil) wrote:
: In article <3ic4ue$o...@fido.asd.sgi.com>, kn...@knobi.munich.sgi.com (Martin Knoblauch) writes:
: |> In article <3iarfd$c...@ra.nrl.navy.mil>, de...@enterprise.ait.nrl.navy.mil
: |> (Butch Deal) writes:
: |> |> In article <3i5odn$g...@rugch4.chem.rug.nl>, hoe...@chem.rug.nl (Frans
: |> |> van Hoesel) writes:
: |> |> |> Dirk Grunwald (grun...@foobar.cs.colorado.edu) wrote:
: |> |> |>
: |> |> |> : I think a lot of this is based on familiarity & what you've used
: |> |> in
: |> |> |> : the past.
: |> |> |>
: |> |> |>
: |> |> |> This easely to improve on when installing SGI.
: |> |> |> 1) the system comes pre installed
: |> |>
: |> |> yeah with everything in one large partition!
: |> |>
: |>
: |> This is the second time that I see this kind of negative
: |> remark on this particular topic in this thread. What is the
: |> exact problem with the one partition layout? Definitely beats
: |> a small root partition that fills up every second day :-)
: |>
: sure but it doesn't beat a small root partition that DOESN'T fill up.
: There are a few places all over that anyone can write to. When you have
: one partition, what you really have is a large root partition with several
: places anyone can write to. Home directories are in the one partition as well.
: pretty easy till fill up the large partition with your user partitions there
: as well as a few temp dirs.
How do you suggest setting up a small root partition that doesn't fill
up?
--
- Paul "Shag" Walmsley <ccs...@cclabs.missouri.edu>
"I'll drink a toast to bold evolution any day!"
Butch Deal
|> I was going to forgo this particular religious war, but ... I'm feeling
|> masochistic ..
jump right in :)
Don't let people write to root areas.
use seperate partitions for areas that people need to be able to write to.
Filesystem kbytes used avail capacity Mounted on
/dev/dsk/c0t3d0s0 28747 11651 14226 45% /
enterprise:/usr 480919 106418 326411 25% /usr
enterprise:/export/exec/kvm/Solaris_2.4_sparc.sun4c/usr/kvm
96031 26390 60041 31% /usr/kvm
/proc 0 0 0 0% /proc
fd 0 0 0 0% /dev/fd
/dev/dsk/c0t3d0s5 19213 1894 15399 11% /var
swap 75016 8 75008 0% /tmp
/dev/dsk/c0t3d0s3 38281 7937 26524 23% /cache
/cache/cache/.cfs_mnt_points/enterprise:_export_raid1_SunOS_5.4_opt
28799866 21723379 6788489 76% /opt
/cache/cache/.cfs_mnt_points/enterprise:_export_raid1_SunOS_5.4_local
5839484 2253140 3469556 39% /usr/local
/cache/cache/.cfs_mnt_points/enterprise:_export_raid1_SunOS_5.4_openwin
28799866 21723379 6788489 76% /usr/openwin
enterprise:/export/enterprise1/mail
968662 65053 806749 7% /var/mail
brandon f chubb
In article <BZS.95Fe...@world.std.com>,
Barry Shein <b...@world.std.com> wrote:
>
>World.std.com (The World) is an SGI Challenge XL with 8x200MHZ R4400s,
>512MB, 5xSCSI (3 F+W, 2 F), etc. We also have an SGI CHallenge L with
>4x100MHZ, 256MB, 3 SCSI, etc as a news server and a FDDI between them.
>
>We're running IRIX 5.3.
At the risk of Barry cutting me off from World..I must put in two
cents from a user perspective. I'm not addressing the admin part of
it at all.
I much preferred when World was on a Solbourne. I mainly run 2 commands
on this system, mail and trn, both of which have what I think are serious
flaws. It makes it a relief to be back on OSF.
The SGI 'mail' utility has a funny habit of skipping every other message
if I'm reading and deleting each one. This is just berkely mail -- nothing
fancy. Works fine on OSF, Sun OS, Solaris, the Intel Unices, and any other
Unix I've tried.
The newsreader trn has a funny habit of not starting at the beginning of
the unread articles, whether just beginning reading a newsgroup or using
the '^' to go to the first article. (This may happen because of my particular
settings, but hey, it happens.)
So on the basis of using mainly two commands and both of them having
problems I'd wonder how much more is broken on SGI.
It may be that its not SGI's fault -- if someone has seen these problems,
please set me straight!
>AIX? You couldn't pay me enough...
-- brandon
Nigel Tzeng
Ben Marcotte <b...@chinook.uoregon.edu> wrote:
>Butch Deal (de...@enterprise.ait.nrl.navy.mil) wrote:
[snip]
>
>(cough cough sputter sputter - pick myself up off the floor)
>
>What color is the sky on your planet?
>
>Solaris - easy, reliable? You can use those three words in one sentence?
>It has been the biggest pain in the *ss for us!
Given the choice a lot of sysadmins I know would rather have 4.1.3 machines
than Solaris machines...
>For me it constantly feels like Sun is working too hard on the toys for the
>OS (Wabi, SDK's, etc...) and not enough on the basics. By basics I mean:
In theory 2.4 is stable. We have 2.3 and it's semi-stable...stable enough
to lull you into a trusting stupor and then crashing on you just before
a major demo.
[snip]
>2) The lp services have a number of problems. The
>/var/spool/lp/system/pstatus file is not correctly formatted. By default,
[snip]
>nobanner="no" to "yes"). We recently had our printer power cycle on us
>which should not have caused too many problems for a fault tolerant OS, but
>instead Solaris let the device drivers in the kernel get corrupted and we
>were forced to reboot the system (and, no, restarting the printer daemons
>did absolutly nothing).
Heh...if it even lets you try to restart the daemon. The official fix
from Sun (I presume it's the official fix since a tech read it off to
me :) is to go into your printer spool directory and delete the contents.
I presume its the same problem you have since it happens to us when our HP
gets cycled when the MIO card gets flakey. The caveat is our lp is set up
to go through another machine (a 4.1.3 machine).
>3) Have you followed any of the threads in comp.unix.solaris on setting
>up a modem for dial-in? I challenge you to find anyone on that group who
>will say that it is easy to "maintain, configure, install, and upgrade" a
>bidirectional modem on a Sparc. I've been trying for months to get ours to
>work under 2.3 and I have not been able to get anything out of it.
I got great working notes regarding this issue when I ordered PPP software
from Morning Star. I actually haven't tried to do dial in yet but there
is a longish section on figuring it out on Solaris.
>4) Lets not forget that Solaris 2.1 was shipped with a completely unusable
>copy of xdm. I call that releasing your beta copy NOT releasing a
>"reliable" product.
Well...so did AIX.
>There are many more problems that a quick glance through the sun
>newsgroups would reveal.
A particularly annoying one here is when Solaris decides not to release
rpcs when programs exit. You have to go in by hand using rpcinfo -d and
delete them. Calling various vendors with somewhat sloppy handing of
rpc resources gets the response "It's a Solaris bug...we don't know what
to do yet but the same code runs on everything else."
[snip]
>: would have to rate then higher than HP's and AIX systems though. DEC's don't
>: really count in the comparison till they get a little closer to UNIX.
> ^^^^^^^^^^^^^^^^^^^^^^^
>Care to explain why OSF/1 does not meet you standards but an OS with the
>problems I listed above does? I tend to think that a system that can't
>even keep track of who is logged in correctly is not a complete Unix!
I don't think that OSF/1 is any less unixy from a users standpoint (we
have one but it's being used as a nice X term at the moment) but our
past porting efforts on the older versions of the OS have been
hellish. Mind you we ported to the alpha a year (more?) ago so things
may be better but code that ported just fine from the sun to hp and
sgi was annoyingly difficult to get running on the alpha.
Hmmm...I'm wondering if that was a port to Ultrix...but I don't think so.
Our Ultrix machine is a Decstation or something or other.
>: Butch Deal de...@ait.nrl.navy.mil
>Ben Marcotte b...@chinook.uoregon.edu (503) 346-4592
brandon f chubb
Michael I. Gold <go...@puck.esd.sgi.com> wrote:
>b...@world.std.com (brandon f chubb) writes:
>
>>The SGI 'mail' utility has a funny habit of skipping every other message
>>if I'm reading and deleting each one. This is just berkely mail -- nothing
>>fancy. Works fine on OSF, Sun OS, Solaris, the Intel Unices, and any other
>>Unix I've tried.
>
>(BSD). /bin/mail is *not* "berkely" mail. Try:
You're right about Mail working better! (Thank you)
But..
world% which mail
/usr/ucb/mail
world%
I would think /usr/ucb/mail would be Berkeley mail. Isn't it?
I'll stick with "BSD" if I ever have to write that again..
-- brandon
cjohn mace
Paul 'Shag' Walmsley <ccs...@sgi2.phlab.missouri.edu> wrote:
>How do you suggest setting up a small root partition that doesn't fill
>up?
No /var, no /tmp and no files or directories with 777 permission, or owned
by anyone other than root/bin/sys/.... Thus, a root filesystem that doesn't
fill up.
>- Paul "Shag" Walmsley <ccs...@cclabs.missouri.edu>
cjohn mace
cjm...@amoco.com
ps-of course quotas would probably stop 99% of the full single partition
systems.
Ade The Shade
: In article <3i5odn$g...@rugch4.chem.rug.nl>, hoe...@chem.rug.nl (Frans van Hoesel) writes:
: |> Dirk Grunwald (grun...@foobar.cs.colorado.edu) wrote:
: |>
: |> 3) installing the other stuff (C++ fortran, nfs etc) takes another
: |> hour (because the cd ain't any faster; not because the work involved
: |> which is nearly nothing.
: make sure you install them in the right order though.
The installation sequence numbers are printed on them.
--
__ __/__ . __ __ | Pope Adrian IV of the Church of The Holy Lungfish,
(_/(_// / (_// / | Larry the Thrice-blessed. Dew de la BAAWA.
The blade of grass that does not bend with the wind
bends the wind around itself.
Michael I. Gold
>The SGI 'mail' utility has a funny habit of skipping every other message
>if I'm reading and deleting each one. This is just berkely mail -- nothing
>fancy. Works fine on OSF, Sun OS, Solaris, the Intel Unices, and any other
>Unix I've tried.
Methinks you're running /bin/mail (SYSV) instead of /usr/sbin/Mail
(BSD). /bin/mail is *not* "berkely" mail. Try:
alias mail Mail
and all will be well.
-- Michael
P.S. /bin/mail is not broken, it just has a different interface with
which you are apparently unfamiliar.
Michael I. Gold
>You're right about Mail working better! (Thank you)
>But..
>world% which mail
>/usr/ucb/mail
>world%
>I would think /usr/ucb/mail would be Berkeley mail. Isn't it?
Well, since /usr/ucb is not a part of IRIX, you'll have to ask Barry
what he put in there. :-)
-- Michael
Roman Fietze
I have joined this thread a little late, but I would add Clones. The
only clones I have worked with are Sun clones from Tatung (which are
better than Suns, IMHO) but Tatung also makes a PREP box which is
basically an RS/6000 clone. I don't know of any HP or SGI clones,
and no one in their right mind would make an Alpha clone.
Sure, and hundreths of companies made 286 clones and loaded them with
MESS-DOS. This is no argument. Decisions what to buy are mostly made
by people who do not have the technical background, or are just some
bean counters who count the initial costs and not the lifetime
costs. I tell you, it's not easy to buy what you think is good when
you are working in a big company. Our company bought hundreths of PC's
with MS-Windows and MS-DOS and I tell you I have and had a hard time
to defend our SCO/HP environment for embedded systems SW development
(sure including textsystem and other comercial stuff and all the nice
things that come with any UNIX like mail, WWW, ...).
Roman
---
Kodak AG Germany fie...@kagcpd01.ag01.kodak.COM
Joakim Rastberg
>Butch Deal (de...@enterprise.ait.nrl.navy.mil) wrote:
>: Suns with Solaris 2.x are about the easyest to maintain, configure, install,
>: upgrade, etc, etc. and are about the most reliable.
>Not a flame by any means, but at the start of January our brand new SPARC
>1000 came online (it arrived in August and they spent the time inbetween
>updating/customizing it). It has 1/4 GB of real memory, 2 CPUs, and a
>"storage array". It ONLY handles our email (PINE) users (ie. no compilers,
>stats, database, etc. software - JUST email).
>The machine will grind to almost a halt at between 60-80 users - but it can
>run okay with 120 users - but only for a while. It was running Solaris 2.3
>in January, but they upgraded to 2.4 in an attempt at "fixing up" the problem
>we are having. 2.4 did not fix up anything.
Do you by any chance have SUN's DECNET solution installed? That sorry
excuse for software can make any Sol2-machine into a worthless piece
of 8-times-a-day crashing junk!
Speaking from experience.
joakim
-------
Joakim (j...@xinit.se) Rastberg, Xinit AB, Sjogatan 2, 852 34 Sundsvall, Sweden
We are Pentium of Borg. Division is futile. You will be approximated.
Wilf Lee
>what i originally wanted to know also:
>
>1. are "standard" machines (pc's) capable to keep track with
> proprietary risc machines (servers - workstations).
Not a chance as far as up time is concerned. The HP9000 is solid and have
been up for more than a year without any problems. The Pentium 60 have been
up and down at least twice a month. For the $$ amount we saved on hardware,
we paid it right back on down time.
>2. is it worthful to bother about configuring unix if
> winnt sets up without doing anything "manually" (nearly ;) )?
Even though NT intalls easily, it doesn't mean it will run optimally with
the default install configuration. In the end, you spend the same time
with unix and NT. I'd stay with unix for it's better supported with more
software and cheaper overall.
>3. is the difference in ease of administration (if this is true)
> and efficiency worth the price difference ?
Unix is better for the simple reason it's been available longer so better
tools are available. As for NT, give it another 3 years to catch up.
>4. whats about software for the different plattforms
> (ease of use - stable - price - public domain)
Except for ease of use, this is definitely where unix wins. Unix is more
stable, cheaper and with more software available. After a year of waiting
and listening to MS FUD, we gave up on Microsoft Exchange Server and went
with HP OpenMail.
regards,
WL
Disclaimer: views expressed are mine and do not represent my employer.
All other standard disclaimers apply.
Barry Shein
From: go...@puck.esd.sgi.com (Michael I. Gold)
>>I would think /usr/ucb/mail would be Berkeley mail. Isn't it?
>
>Well, since /usr/ucb is not a part of IRIX, you'll have to ask Barry
>what he put in there. :-)
It's just a symlink to /usr/bsd/ to ease the transition.
Martin Knoblauch
|> >>I would think /usr/ucb/mail would be Berkeley mail. Isn't it?
|> >
|> >Well, since /usr/ucb is not a part of IRIX, you'll have to ask Barry
|> >what he put in there. :-)
|>
|> It's just a symlink to /usr/bsd/ to ease the transition.
|>
does "/usr/bsd/mail" come from on your system :-)
Butch Deal
|> In article <3id92a$l...@news.missouri.edu>,
|> Paul 'Shag' Walmsley <ccs...@sgi2.phlab.missouri.edu> wrote:
|>
|> >How do you suggest setting up a small root partition that doesn't fill
|> >up?
|>
|> No /var, no /tmp and no files or directories with 777 permission, or owned
|> by anyone other than root/bin/sys/.... Thus, a root filesystem that doesn't
|> fill up.
|>
|> systems.
|>
it would have to be restrictive to prevent two or more users from
inadvertantly filling it up.
Nathan F. Janette
writes:
> Wayne, have you worked with NeXTStep (I found this note in a NeXTStep
> newsgroup), and could you give me a comparison with linux? I'm trying to
> set up a NeXTStep-based UNIX network for a high school district. Trying
> to get NeXTStep to do ANYTHING I want has been like pulling teeth.
Could you please provide some details with this flame?
I've found NEXTSTEP to be the easiest Unix to install
and configure without exception in my experience
(HP-UX, Irix, etc). The Admin GUI tools are very good
in the current revision. Where NEXTSTEP falls short is
in keeping the BSD unix utilities up to date, but that
hasn't affected the generic install/config process in
my experience. There are always things I would like
to be better about NEXTSTEP, but isn't that true for
all our favorite OSs?
Cheers,
-Nathan
--
Nathan Janette
Systems Manager, Axel T. Br nger Lab
Internet: nat...@laplace.csb.yale.edu
Voice: 203 432 5065
Fax: 203 432 3923
Allan Schaffer
>
>The SGI 'mail' utility has a funny habit of skipping every other message
>if I'm reading and deleting each one. This is just berkely mail -- nothing
>fancy. Works fine on OSF, Sun OS, Solaris, the Intel Unices, and any other
>Unix I've tried.
In a former life I developed a taste for the BSD 'mail' program as
well, & use "capital M" 'Mail' (not lowercase 'mail') on the SGI's.
I haven't noticed any "skipping".
MAIL_BSD(1) Mail - send and receive mail
MAIL_ATT(1) mail - send mail to users or read mail
^
To be truly honest, though, nowadays I use zmail. It can run in both
a GUI-mode and a text-only mode.
> The newsreader trn has a funny habit of not starting at the beginning
> of the unread articles, whether just beginning reading a newsgroup or
> using the '^' to go to the first article. (This may happen because
> of my particular settings, but hey, it happens.)
I use & prefer trn as well, but this problem does not happen to me.
My guess is that 'world' has an old version or as you said, something
wierd in your settings.
Allan
wow, this is going to a lot of unrelated newsgroups. Don't know where
you're reading from, so...
--
Allan Schaffer
Silicon Graphics
asch...@sgi.com
http://reality.sgi.com/employees/aschaffe
Charlie McGuire
Woan) writes:
|> In article <3i75o9$1...@info.epfl.ch>,
|> Stefan Monnier <mon...@di.epfl.ch> wrote:
|> >Well, AIX isn't the only unix with an advanced file system.
|> >It's actually getting common to offer a log-structured FS addtionnally
|> >to plain UFS. It's so much faster to reboot and to repartition !
|>
|> IBM was the first of the big players to offer this (contributed this
|> to OSF) and in the latest rev, we also offer inline compression along
|> with some more tuning parameters.
|> --
|> +------All Views Expressed Are My Own And Not Necessarily Shared By
|> IBM-----+
Simply having a Journaled filesystem is a good thing, but IBM's logical
volume manager takes everything one major step forward. The LVM is
probably one of the biggest time/resource savers I've seen in any "Unix"
system. No more staying up late at night to repartition some silly disk
or create new filesystems, or extend live ones. All with no downtime.
The rumors I've heard are that DEC is giving up on supplying anything
like the Logical Volume Management.
Charlie
--
*************************************************************************
( Charlie McGuire | Tel. : (406) 243-4618 )
( System Administration | Fax : (406) 243-4076 )
( The University of Montana | )
( Computer Science Dept | E-Mail: mcg...@cs.umt.edu )
*************************************************************************
Peter Shenkin
Martin Knoblauch <kn...@knobi.munich.sgi.com> wrote:
> This is the second time that I see this kind of negative
>remark on this particular topic in this thread. What is the
>exact problem with the one partition layout? Definitely beats
>a small root partition that fills up every second day :-)
Very simple: it's not what people know, are used to, and love to
hate. That makes it bad. It's a gratuitous change that serves
no real purpose. If a small root partition fills up daily, make
the root partition bigger.
-P.
--
************************ The secret of life: *************************
*Peter S. Shenkin, Box 768 Havemeyer Hall, Chemistry, Columbia Univ.,*
* New York, NY 10027; she...@columbia.edu; (212) 854-5143 *
************* If you find a loose thread, don't pull it. *************
Butch Deal
|> In article <3ic4ue$o...@fido.asd.sgi.com>,
|> Martin Knoblauch <kn...@knobi.munich.sgi.com> wrote:
|>
|> > This is the second time that I see this kind of negative
|> >remark on this particular topic in this thread. What is the
|> >exact problem with the one partition layout? Definitely beats
|> >a small root partition that fills up every second day :-)
|>
|> Very simple: it's not what people know, are used to, and love to
|> hate. That makes it bad. It's a gratuitous change that serves
|> no real purpose. If a small root partition fills up daily, make
|> the root partition bigger.
|>
The problem is that if you put /usr and all the tmp dirs. all in one large
partition, how do you keep it from filling up. if you have a nice small root
partition with seperate partitions or links for all the write-able areas
the root area will not fill up. There is no reason to hate seperate root
partitions.
Steve Strange
> Simply having a Journaled filesystem is a good thing, but IBM's logical
> volume manager takes everything one major step forward. The LVM is
> probably one of the biggest time/resource savers I've seen in any "Unix"
> system. No more staying up late at night to repartition some silly disk
> or create new filesystems, or extend live ones. All with no downtime.
> The rumors I've heard are that DEC is giving up on supplying anything
> like the Logical Volume Management.
The rumors you heard were wrong. DEC OSF/1 currently offers LSM,
which is very similar in functionality to IBM's LVM. But what might
be more interesting to you is that Digital's log-based file system,
AdvFS, offers you the management features you talk of here without
needing LVM or LSM! An AdvFS domain can be created from one or more
disk partitions, and any number of filesets can be created in that
domain and mounted as you would mount a UFS partition. New filesets
can be added at any time and mounted, or existing ones unmounted and
deleted. You can then add and remove volumes (partitions) to and from
domains to increase or decrease their size, all while the filesets in
that domain are live and accessible. Note that you can move domains
from disk to disk in this fashion, by adding partitions on the new
disk into a domain, then deleting the partitions from the old one.
There's even a nice GUI interface to do it all. So not only has DEC
not given up on this type of functionalty, they've had it for quite a
while now!
Sorry for sounding like an advertisement, but I just want to set the
record straight.
Steve
--
Steve Strange
Digital Equipment Corporation, Nashua, NH
DCE DFS for DEC OSF/1 -- Development
str...@zk3.dec.com
Paul Jackson
|>
|> In article <3ic4ue$o...@fido.asd.sgi.com>,
|> Martin Knoblauch <kn...@knobi.munich.sgi.com> wrote:
|>
|> > What is the exact problem with the one partition layout?
|>
|> hate. That makes it bad. It's a gratuitous change that serves
|> no real purpose.
Thanks, Peter, for your direct answer to Martin's question.
I am aware of one good purpose that defaulting to a single
root+usr file system serves us. We hear quite a few reports of
customers having difficulty installing into the separate small
root partitions, because they lack enough free space for the
second copy of /unix (unix.install or whatever). This often shows
up in the miniroot, which is for many a most inhospitable
environment for managing disk space. And just significantly
increasing the root partition isn't a pure win either, because
then less skilled or energetic users will be more likely to waste
a significant chunk of disk, as they fill up one partition while
the other lies underutilized.
"It's a gratuitous change" ...
By the same token, changing it back would be gratuitous as well,
right? So we should leave it be, right? <g>
--
I won't rest till it's the best ...
Software Production Engineer
Paul Jackson (p...@sgi.com), x1373
Norman Ma
I am very pleased with SGI from a sys. admin point of view. I
am also very happy w/ SGI's developer's option; everything is
there.
HP is a fast workstation, but they need to spend more effort in
packaging the software and hardware into a workable system.
The same for Sun. It seems like that the necessary parts are
scattered through half of the United States. I can almost
hear the Sun configuration people say, "Hey, we are Sun, don't
everyone know about our computer?"
--Norman Ma
Peter Shenkin
Paul Jackson <p...@sam.engr.sgi.com> wrote:
>In article <3ig2uc$b...@sol.ctr.columbia.edu>, she...@still3.chem.columbia.edu (Peter Shenkin) writes:
>|> NNTP-Posting-Host: still3.chem.columbia.edu
>|>
>|> In article <3ic4ue$o...@fido.asd.sgi.com>,
>|> Martin Knoblauch <kn...@knobi.munich.sgi.com> wrote:
>|>
>|> > What is the exact problem with the one partition layout?
>|>
>|> Very simple: it's not what people know, are used to, and love to
>|> hate. That makes it bad. It's a gratuitous change that serves
>|> no real purpose.
>
>Thanks, Peter, for your direct answer to Martin's question.
>
>I am aware of one good purpose that defaulting to a single
>root+usr file system serves us. We hear quite a few reports of
>customers having difficulty installing into the separate small
>root partitions, because they lack enough free space for the
>.... And just significantly
>increasing the root partition isn't a pure win either....
There are no pure wins; both "solutions" have problems. Since
we know what problems to expect when / and /usr are separate file
systems, and are used to dealing with them, and since having separate
file systems is the "norm", a change from this seems gratuitous
to me, unless the change is REALLY A BIG WIN. It's not enough
for it to be better. It has to be A HELLUVA LOT BETTER. It's
hard for me to believe that it is, though I am possibly wrong.
It reminds me a bit of what IBM did with AIX. They made many,
many changes which make any UNIX administrator not used to their
particular platform feel, well, not at home. It is quite possible
that these changes, taken quite objectively, are each for the better.
But their betterness, even if assumed, is not, IMHO, and in the HO of
many others, sufficient to warrent the deviation from general practice
and from what one expects when one sits down at a putative UNIX machine.
> "It's a gratuitous change" ...
>By the same token, changing it back would be gratuitous as well,
>right? So we should leave it be, right? <g>
At this point, perhaps so. 0.5*<g> Just don't do it again. 0.25*<g>
Jim Sullivan
>In <3ib5th$2...@osiris.wu-wien.ac.at>, Rupert Thurner <rthu...@x11srv.edvz.uni-klu.ac.at> writes:
>>what i originally wanted to know also:
>>
>>1. are "standard" machines (pc's) capable to keep track with
>> proprietary risc machines (servers - workstations).
>
>Not a chance as far as up time is concerned. The HP9000 is solid and have
>been up for more than a year without any problems. The Pentium 60 have been
>up and down at least twice a month. For the $$ amount we saved on hardware,
>we paid it right back on down time.
Not that I want to get into an uptime contest...but who was the manufacturer
of the Pentium? Often times I see these type of comparisons between
major, non-intel OEM's and some second or third tier Intel compatible
supplier. Hardly a fair comparison, in my opinion. Of course, the
operating system is part of the equation as well, but you didn't indicate
which OS was in use.
--
Jim Sullivan
SCO Canada, Inc. Toronto, Ontario "Lawyers are for people too gutless
j...@sco.com, ...!uunet!scocan!jim to shoot. I'd rather be shot"
416 960 4042 (922 8397 for fax) -- John Perry Barlow @ SCO Forum 94
--
Jim Sullivan
Ken Green
> Simply having a Journaled filesystem is a good thing, but IBM's logical
> volume manager takes everything one major step forward. The LVM is
> probably one of the biggest time/resource savers I've seen in any "Unix"
> system. No more staying up late at night to repartition some silly disk
> or create new filesystems, or extend live ones. All with no downtime.
> The rumors I've heard are that DEC is giving up on supplying anything
> like the Logical Volume Management.
I thought LVM was part of OSF/1. Certainly the version that ships with HP-UX
is decended from the OSF version. Since DEC uses OSF as their Unix
implementation why would they ant to drop this technology ?
--
__________________email Ken....@kgcc.demon.co.uk _____________________
Ken Green Computer Consultancy
22 Matthews Chase, Binfield, Berkshire, RG12 5UR. U.K.
Mark Dapoz
> The LVM is
>probably one of the biggest time/resource savers I've seen in any "Unix"
>system. No more staying up late at night to repartition some silly disk
>or create new filesystems, or extend live ones. All with no downtime.
If you want real flexibility in a volume manager then you should take
a look at AFS (Andrew Filesystem) from Transarc. It makes the features
provided by LVM look primitive in comparison. Of course using AFS just
because of its volume management abilities is a bit of overkill since
it provides so many other benefits.
--
Mark Dapoz (m...@gated.cornell.edu)
Art Stine
The rumors I've heard are that DEC is giving up on supplying anything
like the Logical Volume Management.
I think you've heard wrong... they already supply it (AdvFS) has been
supplied for quite awhile now on OSF/1.
-art stine
visix software
Peter da Silva
Butch Deal <de...@ait.nrl.navy.mil> wrote:
> |> familiar with BSD. Solaris isn't either. OSF/1 is pretty straight BSD with
> |> a little System V admin stuff on the edge.
> just enough to make it a pain in the...
I'm not sure where you're coming from with this. Virtually all the important
interfaces are BSD, except they dropped the version-7-ish BSD /etc/rc stuff
and went to inittab. I was a little annoyed they put /etc/rc*.d in /sbin
but that's pretty minor.
Perhaps you're trying to think of it as a System V box. It's not. It's a
BSD box with a few System V features.
> this just makes it neither. You can't sit on the BSD/SYSV fence
> forever.
I don't think they are. They've pulled in some good features from System V
but kept the basic BSD interfaces. Solaris has pretty much sucked in the
bad (like the printer configuration) with the good (like inittab).
> It seems that all the people that do well with BSD just have a reall hard time
> with solaris 2.x. Solaris 2.x is one of the closest SYSV OS's.
I have a BSD machine and a System V machine at home. We used to be a complete
System V and Xenix shop. You can't simplify things and just say it's a System
V versus BSD problem. Well, maybe you can but it's not true.
And then there's the Solaris things that are neither SysV nor BSD, like the
whole /usr/ccs tree.
--
Peter da Silva `-_-'
Network Management Technology Incorporated 'U`
1601 Industrial Blvd. Sugar Land, TX 77478 USA
+1 713 274 5180 "Hast du Heute schon deinen Wolf umarmt?"
Peter da Silva
Will Morse <wi...@Starbase.NeoSoft.COM> wrote:
> they can be _just_as_good_enough_ and even cheaper. DEC is, at least
> at present, NOT good enough and certainly not cheap enough. This
> gives clone makers no where to go.
In what way is DEC "not good enough"? Their machines are blazing fast, their
UNIX is solid, and the only downside is they're pretty pricey.
Dave Olson
| Another example: I had trouble with the SGI's default malloc that is
| build into libc. There is a `malloc (3X)' package in a seperate
| library and that is obviously better. If I inteprete SGI's behaviour
Why is it "obviously better"? Different, yes. Better for some apps,
yes (worse for others). More knobs to turn, definitely ;)
| right there is customer code that relies on some undocumented feature
| of the old malloc, so SGI didn't exchange the old code. I think Sun
| would have done the other way and force the customer to change its
| code. Again, I feel better with Sun's way.
Ahh, that comment about allowing the use of a pointer after free.
That comment is straight out of the original libmalloc (from bsd 2.8
or 4.1, or thereabouts). Believe it or not, there is still code that
counts on that behavior (from v7 unix). *Neither* of our standard malloc's are
the same as the "original" libc or libmalloc malloc's, but we do allow
that usage to still work in the libc malloc.
I imagine you will find the Sun libc still allows this also, even if
it doesn't document it.
| I can understand why people hate Sun, it is possible to miss deadlines
| because of changes in Suns environment. Howver, I prefer to do the
| right thing as soon as possible (I'm can type fast :-).
I think you have a somewhat unusual idea of "right thing", but you
are entitled to believe whatever you want.
| Both Sun and SGI have less floating point power than SGI/DEC/IBM.
Hmm; less floating point than ourselves, well yes, that's true, for different
classes of our machines, but it is an interesting comment ;)
--
The most beautiful things in the world are | Dave Olson
those from which all excess weight has been | Silicon Graphics
removed. -Henry Ford | ol...@sgi.com
Russell Street
>|> no real purpose. If a small root partition fills up daily, make
>|> the root partition bigger.
>The problem is that if you put /usr and all the tmp dirs. all in one large
>partition, how do you keep it from filling up. if you have a nice small root
>partition with seperate partitions or links for all the write-able areas
>the root area will not fill up. There is no reason to hate seperate root
>partitions.
Then what do you do about your or your operators typing
tar -cf /dev/mt/tsp0d6 / ...
or something similiar?
Russell
Dave Olson
| Very simple: it's not what people know, are used to, and love to
| hate. That makes it bad. It's a gratuitous change that serves
| no real purpose. If a small root partition fills up daily, make
| the root partition bigger.
It wasn't a gratuitous change (since I was one of the 3 people that made the
decision, I think I have a fair amount of insight into this!).
As others have said, the support folks got a *lot* fo calls from
"naive" users who didn't understand why they ran out of disk space when
they had lots of spare disk (either tmpfile space, or because they kept
copying stuff to / or files on /, or whatever). At the same time, we
were starting to ship Indy, which was targeted in part, at folks who
were, and *wanted to remain*, naive, in the computer admin sense.
By the way, it's not all that new. Most of the 11/60's and 11/70's I
used, running V6, V7, and BSD 2.8 unix used the entire "system disk"
as a single partition, except for the swap area. It's all in what
works for your environment, and what you are used to.
All the arguments about keeping root 'pure' are just plain silly.
There's nothing sacred about not running out of space on / vs /usr
both have their problems.
Barry Shein
From: pe...@nmti.com (Peter da Silva)
>In article <3i864q$8...@starbase.neosoft.com>,
>Will Morse <wi...@Starbase.NeoSoft.COM> wrote:
>> they can be _just_as_good_enough_ and even cheaper. DEC is, at least
>> at present, NOT good enough and certainly not cheap enough. This
>> gives clone makers no where to go.
>
>In what way is DEC "not good enough"? Their machines are blazing fast, their
>UNIX is solid, and the only downside is they're pretty pricey.
I think what he meant was that, as I express it, "good enough" tends
to win.
That is, people won't pay much of a premium for a little or even quite
a bit better if what they can get for less gets what they want done.
So it's "good enough" in that sense, good enough for the job, so I
think what he's saying is, to bring it back into context, you want to
clone systems that are high volume because they're in that "good
enough" category. It's the same reason people buy a lot more honda
civics and toyota corollas and chevys and fords than they do much
better engineered cars; they're good enough (to get them to work and
the shopping etc and back), so why spend $25K-$50K when you can spend
$15K unless you have a special need or some personal obsession with
quality?
When computers all pretty much sucked in performance, performance was
something to get at any price you could manage, even if it stung the
budget. But if you can get a 100MIPS box for $5000 (with disk etc) and
you're not particularly compute-intensive in your uses how much more
would you pay for 150MIPS or 250MIPS? Not much, maybe not anything,
you're happy with what you get for $5000, maybe you'd rather have more
disk or a nicer monitor or something instead of spending on the MIPS.
This is a lot of the reason (IMHBCO) that DOS and Windows does so well
in the marketplace, it sucks (IMHBCO), but for what many people want
to do it's Good Enough (tm), stick it on a 486DX66 with a 540MB IDE
drive and 16MB of memory with a 15" monitor for like $2500 or less,
with some software even, and away you go. And for a large part of the
market a 486DX33 with 4MB and a 340MB disk and a 14" monitor is "good
enough", for barely over $1000 (probably under.)
Every market has its own "good enough" level.
Some markets, like large servers, compute-intensive and serious
graphics (I suppose a form of compute-intensive) aren't anywhere near
their "good enough" level so will still spend a premium on the
absolute best they can manage, mortgage the house, send the kids off
to work, rent out the spouse, we've got multi-media to produce!
And the clones will tend to follow the "good enough" curve because
that's where they expect the volume to be. They're only going to grab
some small percentage of the market, particularly in the beginning, so
may as well grab a small piece of a bigger pie. The resistance to
buying clones is probably less in their specific price/performance
(tho that has to be better than The Real Thing) and more in losing
because of lack of service, sales and distribution channels, name
brand anxiety, fear that they'll vaporize or their hardware will
become obsoleted or be subtly incompatible, etc.
Once it can compute everything you need in the vertical blanking
interval you're done.
Barry Shein
From: kn...@knobi.munich.sgi.com (Martin Knoblauch)
>In article <BZS.95Fe...@world.std.com>, b...@world.std.com (Barry
>Shein) writes:
>|>
>|> From: go...@puck.esd.sgi.com (Michael I. Gold)
>|> >>I would think /usr/ucb/mail would be Berkeley mail. Isn't it?
>|> >
>|> >Well, since /usr/ucb is not a part of IRIX, you'll have to ask Barry
>|> >what he put in there. :-)
>|>
>|> It's just a symlink to /usr/bsd/ to ease the transition.
>|>
>
> Which has no "mail" in it, at least under IRIX-5.2/5.3. So, where
>does "/usr/bsd/mail" come from on your system :-)
/usr/bsd/mail is a symlink to /usr/bsd/Mail, we try to steer people
away from mail_att unless they work at it.
Hamish Marson
> > Simply having a Journaled filesystem is a good thing, but IBM's logical
> > volume manager takes everything one major step forward. The LVM is
> > probably one of the biggest time/resource savers I've seen in any "Unix"
> > system. No more staying up late at night to repartition some silly disk
> > or create new filesystems, or extend live ones. All with no downtime.
> > The rumors I've heard are that DEC is giving up on supplying anything
> > like the Logical Volume Management.
DEC told us they were dropping it at last years Uniforum NZ. (May?)
> I thought LVM was part of OSF/1. Certainly the version that ships with HP-UX
> is decended from the OSF version. Since DEC uses OSF as their Unix
> implementation why would they ant to drop this technology ?
It is part of OSF/1 (The reference OSF/1 that is). DEC are dropping
LVM becasue they were either unable or unwilling to integrate into
their version of OSF/1. They have instead supplied a piece of software
called LSM (Logical Storage Manager) which looks just as clunky and
tacked on as their implementation of LVM did.
While IBM's LVM is an integral and smooth part of the operating system,
both LVM and LSM look (From the times I have played with it), like a
poor attempt at copying a well thought out idea. Is LSM going to be dropped
as well? Or are they keeping it?
An Aside...
My favourite DEC release was when they showed the Alpha 2100's, and
asked one member of the audience to pull out any HD from the integrated
cabinet. The system immediately bit the dust. The really foolish part
of it was when the DEC droids asked him to try again....
--
======================================================================
| Hamish Marson |
| Systems Programmer | |
| Computer Services | INTERNET h.ma...@waikato.ac.nz |
| University of Waikato | PHONE +64 7 8562889 xt 8181 |
| New Zealand | FAX +64 7 8384066 |
===========Disclaimer :- Remember. You heard it here first.===========
Ian G Batten
Nigel Tzeng <ni...@access.digex.net> wrote:
> >Solaris - easy, reliable? You can use those three words in one sentence?
> >It has been the biggest pain in the *ss for us!
>
> Given the choice a lot of sysadmins I know would rather have 4.1.3 machines
> than Solaris machines...
Count us as not agreeing. I've got some years of Sun experience (dating
back to SunOS 3.0) and a wide range of other Unix platforms, so I regard
my opinion reasonably considered. Have around 100 Suns hung off an
Auspex and Solaris is decidedly preferred by the three of us that run
the machines. We are trying to wean the last few 4.1.X users over.
Why? Auto install. Cachefs. SMP. Loads of things.
Yes, Solaris is buggy. But those bugs are in implementation and can be
(and are being) fixed. 4.1.X is reliable, but its problems are
architectural and not so easily fixed.
> In theory 2.4 is stable. We have 2.3 and it's semi-stable...stable enough
> to lull you into a trusting stupor and then crashing on you just before
> a major demo.
We have only one unreliable 2.3 machine, a rather elderly 690/512 with a
mess of odd hardware in it, IPI disks on two controllers and ODS. Our
more standard machines are rock solid. 2.3 + all recommended patches.
Our 10/514 and 20/514 machines are very impressive.
> gets cycled when the MIO card gets flakey. The caveat is our lp is set up
> to go through another machine (a 4.1.3 machine).
Our spoolers point at the lpd on our Auspex and lpshut and lpshed work
fine.
> >4) Lets not forget that Solaris 2.1 was shipped with a completely unusable
> >copy of xdm. I call that releasing your beta copy NOT releasing a
> >"reliable" product.
As was 4.0.
ian
Frank Kraemer
|>
|> If a 20 processor SS2000 isn't big enough the Cray 6400 (I think that is the
|> model) goes to 64 SPARC processors on 4 XDbus instead of 2 running
|> essentially the same O/S.
There is a song from Supertramp called DREAMER - listen to it and
think of a running SUN 2000 with 20 CPUs........Solaris can not handle-
8 CPUs right so 20 is a real dream....
-frank-
+----------------------------------------------------------------------+
| E-Mail : krae...@vnet.ibm.com / kr...@ibm.de |
| Voice : +49-(0)611/776-364 / Fax: +49-(0)611/776-500 |
| Mail : Abraham-Lincoln Street 26, 65189 Wiesbaden, Germany |
+----------------------------------------------------------------------+
Pat Wilson
Of course, you want to have an OS locally, too, which is where AIX
comes in... AFS + AIX is a *very* nice combination - never newfs again!
Unfortuantely, there are all those _other_ OSes that handle disks in
primitive ways...
Imagine having to guess how big a partition is going to want to be
when you format a disk! What do those folks use - crystal balls?
:->
--
Pat Wilson
p...@coos.dartmouth.edu
Charlie McGuire
writes:
The only problem with that is $$$$$.
Butch Deal
for just cutting tapes, users would not be able to create the /dev/mt
directory or the /dev/mt/tsp0d6 file. I tend not to do things myself as root
unless root access is needed, as well.
Ben Marcotte
: In article <3i864q$8...@starbase.neosoft.com>,
Why does everyone keep saying that Dec's are expensive? Both of the
Alphastation 200's (4/166 & 4/233) can outperform the entire Sparcstation
20 (single processor) line. The last time I checked, the Alphastation
200's listed at $10295 and $15595 (I will acknowledge that these prices do
not include a monitor). The SS-20 that is the closest in performance (to
the 4/166) is the model HS11 which listed at $18695 (when I check on Nov
8, 94 - this might have dropped).
The closest SGI (again to the 4/166 not the 4/233) would be the
Indigo2 R4400SC which I have an educational discout price of $15925.
Commercial price is probably much higher.
It is hard to compare the RS/6000's because they have low
SPECint's and high SPECfp's. If I had to take a stab at it, I said that
the Model 3AT and 3BT are somewhere around the 4/233 in performce. I
got educational discount prices on these at $18396 and $21200. Again
commercial prices are much higher.
The HP's are also hard to compare because of a lower int/fp ratio.
The HP 715/100 is probably th closesst to the 4/166. It lists at $19005.
The HP 735 ranks slightly below a 4/233 and it lists at $30395.
So basically, to get the equivallent of the $10k Alpha you have to
pay >$15 elsewhere. To get something like a $16k Alpha, you have to pay
commercial prices of >$20k.
==============================================================================
Ben Marcotte b...@chinook.uoregon.edu (503) 346-4592
==============================================================================
Chuck Hedrick
>All the arguments about keeping root 'pure' are just plain silly.
>There's nothing sacred about not running out of space on / vs /usr
>both have their problems.
I think there used to be one good argument: You wanted root to be as
small as possible to minimize the probability that the file system
would be messed up in a crash. Traditionally root had enough files
that a single-user boot used only root. It had the tools needed to
fix the rest of the system. So for the most robust configuration, you
kept it small, and in many cases kept a backup root partition on a
different disk. These days the systems I'm seeing seem to need /usr
in order to come up single user. In that case there seems to be no
real advantage to a separate root. (However I'm inclined to think
that the old idea was good.)
Darryl Trujillo
>>3) Have you followed any of the threads in comp.unix.solaris on setting
>>up a modem for dial-in? I challenge you to find anyone on that group who
>>will say that it is easy to "maintain, configure, install, and upgrade" a
>>bidirectional modem on a Sparc. I've been trying for months to get ours to
>>work under 2.3 and I have not been able to get anything out of it.
>
>Don't use workstations as modem pools or routers. We are in the fortunate
>position to sue de\dicated quipment for all this sorts of tasks.
>Serial ports have always given us trouble on Suns. (ever since SunOS 4.0,
>that is)
>
"Let them eat cake" indeed. This is absurd. Although it's nice to work in a
shop that can afford to buy new hardware all the time (you are indeed
fortunate!), there are a lot of shops with older systems laying around that
make *fine* routers and modem pools, all for less money than your solution
would take.
Serial ports may have always been a problem with Suns, but with Solaris the
problem became a lot worse. I use an old system as a router at a remote site, and
after hassling with Solaris 2.3 serial support for a couple of months, I
installed 4.1 and have been happy ever since. I've had much better luck with
SysVr3 and Linux on PC hardware.
- dmt
Ben Marcotte
: - dmt
Exactly!
Firstly, to respond to Casper. My intent was to set up one host with one
modem on one phone line serving one user at a time. All those ones don't
add up to modem pool or router. But, under Solaris 2.3, one modem is
enough of a problem that as I said, I still can not get it working (even
after using Celeste Stokeley's very informative FAQ guide on this exact
subject).
There are two things that are really frustrating about this situation.
First, the only person that responded that said they had done it, did so
on a machine with an Sbus serial card. I consider this cheating because
it means spending more money upgrading our hardware, to do something it
should have done from the begining, than we spent on the modem!
Second, if I simply took the modem off the sparc and put it on our Linux
box, it would work without any of these problems. This is the ease of
use of Solaris - you learn how easy it is to use other systems! As such,
I would greatly agree with the Linux part of the last line of Darryl's
post (I can't really say anything about SVR3, I haven't used it).
Szymon Sokol
: In article <D4B04...@apertus.com>, ste...@apertus.com (Steve Kappel) writes:
: |>
: |> If a 20 processor SS2000 isn't big enough the Cray 6400 (I think that is the
: |> model) goes to 64 SPARC processors on 4 XDbus instead of 2 running
: |> essentially the same O/S.
: There is a song from Supertramp called DREAMER - listen to it and
: think of a running SUN 2000 with 20 CPUs........Solaris can not handle-
: 8 CPUs right so 20 is a real dream....
Yeah, that is probably true, but the CS64000 (the Cray mentioned above) can
really run with 64 CPUs. Uhm... how many CPUs can be put into IBM RS6000,
any model? What was that saying about a kettle and a frying pan? ;-)
--
Szymon Sokol -- Network Manager
U U M M M M University of Mining and Metallurgy, Computer Center
U U MM MM MM MM ave. Mickiewicza 30, 30-059 Krakow, POLAND
U U M M M M M M M M TEL. +48 12 338100 EXT. 2885 FAX +48 12 338907
UUUU M M M M M M WWW page: http://www.uci.agh.edu.pl/~szymon/