SUMMARY: Sun 690 and Solbourne 700/900

From: David Wheable (djw@advanced-robotics-research-centre.salford.ac.uk)
Date: Sat Mar 07 1992 - 22:30:18 CST


Sorry for the delay in this summary. I think our 4/280 heard I'm thinking
about replacing it because it trashed one of its disks!!

Thanks to everyone who replied. I've included all the messages apart from
those that wished to remain anonymous.

The general feeling is that the Solbourne is a more stable product as it
has had multi-processors for a couple of years now. It also has symetric
multiprocessing working in the current version of the operating system
whereas Sun hasn't. However, when Solaris 2.0 arrives they will both be
using the same SMP code.

On the hardware side, the Solbourne seems to be the more expandable product.
As it has the 128Mb/s Kbus individual processor cards can be plugged into
the backplane, whereas Sun still uses VME and so have put all the processors
on one master board.

On the support side it appears to depend on how good the people in your area
are. Some people have said Sun is better, others Solbourne. What can I say?

Personally, I think the Solbourne has won on the grounds of stability, and
upgrade path.

Thanks to ......
jayl@uucp.bit (Jay Lessert)
texsun.Central.Sun.com!cyrix!vixen!brian (Brian Holgate)
rcd@fed.frb.gov
"Anthony A. Datri" <datri@com.convex.concave>
moore@gov.nih.nlm (Robin Moore)
liz@edu.ucar.cgd.isis (Liz Coolbaugh)
bws900@au.edu.anu.cscgpo (Bede W P Seymour)
Dieter Muller <dworkin@com.rootgroup.merlin>
jimh@com.fmc.nsd (Jim Hendrickson x7348 M233)
paul@uucp.hydres
matt@com.xerox.wbst845e (Matt Goheen)
Billy Barron - VAX/UNIX Systems Manager <billy@edu.unt.acs.sol>
jeff%auratek@NET.UU.uunet (Jeff Martin)
mdl@com.cypress (J. Matt Landrum)
oran@com.amdahl.spg (Oran Davis)
era@edu.ucar.ncar (Ed Arnold)
cmorrow@uucp.tmsl (Chris Morrow)
grover@com.hac.vulcan1 (Dean Grover)
Mike Raffety <miker@com.sbcoc>
"Marc Ph. A. J. St.-Gil - Solbourne Administrator"
peterg@com.murphy.jupiter (Peter Gutmann)
poffen@com.slb.ate.sj (Russ Poffenberger)
hedrick@edu.RUTGERS.dartagnan (Charles Hedrick)
iain@uucp.uksol (Iain Campbell - Rest of Europe)
its@com.Solbourne (Ty Sell)
rodo@com.auspex (Rod Livingood)

David Wheable,
National Advanced Robotics Research Centre,
Salford,
England.

----- Begin Included Message -----

>From jayl%bit%edu.ogi.cse@edu.ogi.cse Tue Feb 25 19:00:08 1992
Received: from nsf.ac.uk by arrc.salf.ac.uk; Tue, 25 Feb 92 19:00:02 GMT
Received: from cse.ogi.edu by sun2.nsfnet-relay.ac.uk with SMTP inbound
          id <20177-0@sun2.nsfnet-relay.ac.uk>; Tue, 25 Feb 1992 18:23:53 +0000
Received: by cse.ogi.edu (5.61+eap+OGI_1.1.named/IDA-1.2.8+OGI_1.12) id AA24915;
          Tue, 25 Feb 92 10:22:41 -0800
Received: by bit.bit.uucp (4.0/SMI-3.2) id AA03259; Tue, 25 Feb 92 10:13:18 PST
Date: Tue, 25 Feb 92 10:13:18 PST
From: jayl@uucp.bit (Jay Lessert)
Message-Id: <9202251813.AA03259@bit.bit.uucp>
To: djw@uk.ac.salford.advanced-robotics-research-centre
Subject: Re: Sun 690 and Solbourne 700/900
Sender: jayl%bit%edu.ogi.cse@edu.ogi.cse
Status: RO

>From: David Wheable <ogicse!advanced-robotics-research-centre.salford.ac.uk!djw>
>Date: Tue, 25 Feb 92 10:06:57 GMT
>To: ogicse!eecs.nwu.edu!sun-managers
>Subject: Sun 690 and Solbourne 700/900
>
>I'm currently looking into upgrading our 4/280 to either a Sun 670 or 690 or
>a Solbourne 700 or 900 Series.
>
>Following the "Calling 690 Managers" messages I'm wondering if the Solborne
>is the better product. I'd be interested in the opinions and experiences of
>this mail group.

Whew, what a host name! :-)

We've had a 3-cpu, 256MB Solbourne 5/600 for over two years now, running
symmetric MP for over a year.

They are industrial-strength, very reliable machines. We beat the hell
out of ours and it keeps going, whether we're running a load average of
20+ small jobs or 10+ big (25MB-500MB) jobs. We use it as a compute/NFS
server, not a multi-user system, so I can't give you any feedback on 100's
of users type usage.

We recently judged the 5E/904 against a 4/690-4 and didn't even have to try
benchmarking the 690, Solbourne was the easy choice for us. A few of the
reasons:

- We needed a production-ready machine now. Sun doesn't have symmetric MP
  yet. Sun not able to make a convincing argument that they'll have any easier
  time debugging SMP than anybody else in the business (it's tough). Plan on
  either doing w/o SMP, or going through 3-6 months of hell when they switch
  over. *And* you get to simultaneously switch to Solaris 2.0. Oh, joy.
  Point to Solbourne.

- Sun best offer price about 10% lower than Solbourne. Point to Sun.

- Sun can/will not sell SCSI boot disk, must have IPI boot disk. Twice as
  expensive and same speed as SCSI. Point to Solbourne.

- When Viking (finally) comes out, you get to upgrade your 690 by removing four
  Cypress CPU's and replacing them with.... TWO Viking's. I'm not making this
  up. Because of Sun's shortsighted econobox packaging decisions (MBus, single
  master board with CPU's on daughter boards), they simply do not have room for
  more than two Vikings (and attendant 2MB cache each) on the 690 master
  board. Essentially zero upgrade path. Solbourne has 9 full-size KBus slots,
  with of room (and backplane speed) for at least one more generation of
  upgrade. Point to Solbourne.

- Application software 100% compatible. We've never found or written an
  incompatible application.

---
Jay Lessert  {decwrl,cse.ogi.edu,sun,verdix}!bit!jayl
Bipolar Integrated Technology, Inc.
503-629-5490  (fax)503-690-1498

>From texsun.Central.Sun.com!cyrix!vixen!brian Tue Feb 25 19:00:28 1992 Received: from nsf.ac.uk by arrc.salf.ac.uk; Tue, 25 Feb 92 19:00:25 GMT Received: from Sun.com by sun2.nsfnet-relay.ac.uk with SMTP inbound id <20832-0@sun2.nsfnet-relay.ac.uk>; Tue, 25 Feb 1992 18:45:06 +0000 Received: from texsun.Central.Sun.COM by Sun.COM (4.1/SMI-4.1) id AA10851; Tue, 25 Feb 92 10:43:16 PST Received: from cyrix.UUCP by texsun.Central.Sun.COM (4.1/SMI-4.1) id AA25969; Tue, 25 Feb 92 12:43:11 CST Received: from vixen.cyrix.com by cyrix.com (4.1/SMI-4.1) id AA05401; Tue, 25 Feb 92 12:30:35 CST Received: by vixen.cyrix.com (4.1/SMI-4.1) id AA00758; Tue, 25 Feb 92 12:30:35 CST Date: Tue, 25 Feb 92 12:30:35 CST From: texsun.Central.Sun.com!cyrix!vixen!brian (Brian Holgate) Message-Id: <9202251830.AA00758@vixen.cyrix.com> To: djw%advanced-robotics-research-centre.salford.ac.uk%texsun@com.Sun.Central.texs Subject: Re: Sun 690 and Solbourne 700/900 Sender: texsun.Central.Sun.com!cyrix!vixen!brian Status: RO

David,

We are using the Solbourne 5E/900 series machines here, and are very happy with the performance,price, and support. These machines are truley wonderful. They have 5 , 384 meg ram, and most of them have at least 4 of the 3 Gbyte 9 Meg/Sec IPI disks from Seagate. The uptime is much better than my 4/470 (which is a real dog) and they run all the aplications that are written for the suns (in fact we have a logic simulator from ikos systems that was not supported on the Solbourne platform by either Ikos or Solbourne, and the local Solbourne engineer worked over the Christmas holidays to get it working) In short I personally will not be buying any more Sun file servers for some time to come.

Regards,

Brian

----------------------------- -------------------------------------- | | Brian Holgate | | ___ | Systems Manager | | / _ \ | 2703 North Central Expressway | | / / \/ | Richardson, Texas, USA 75080 | | ( ( __ ___ __ /\ __ __ | Email: ...!texsun!cyrix!brian | | \ \ /\ \/ /|V _||\/|\ \/ / | or cyrix!brian@central.sun.com | | \___/\ / |_| |__|/_/\_\ | Phone: (214)-234-8388 Ext. 215 | | Cyrix /_/ Corporation | Fax: (214)-680-2401 | ----------------------------- --------------------------------------

>From m1rcd00%fed@net.UU.uunet Tue Feb 25 19:59:21 1992 Received: from nsf.ac.uk by arrc.salf.ac.uk; Tue, 25 Feb 92 19:59:17 GMT Received: from relay2.UU.net by sun2.nsfnet-relay.ac.uk with SMTP inbound id <21781-0@sun2.nsfnet-relay.ac.uk>; Tue, 25 Feb 1992 19:13:35 +0000 Received: from uunet.uu.net (via LOCALHOST.UU.NET) by relay2.UU.NET with SMTP (5.61/UUNET-internet-primary) id AA28989; Tue, 25 Feb 92 13:45:55 -0500 Received: from fed.UUCP by uunet.uu.net with UUCP/RMAIL (queueing-rmail) id 134435.9413; Tue, 25 Feb 1992 13:44:35 EST Received: from cohoes.FRB.GOV by fed.FRB.GOV (4.1/SMI-4.0) id AA02223; Tue, 25 Feb 92 13:31:30 GMT Received: from localhost by cohoes.FRB.GOV (4.1/SMI-4.0) id AA06620; Tue, 25 Feb 92 13:32:14 EST Message-Id: <9202251832.AA06620@cohoes.FRB.GOV> To: David Wheable <djw%advanced-robotics-research-centre.salford.ac.uk%uunet@net.UU.uunet> Cc: rcd%cohoes@net.UU.uunet, erf%cohoes@net.UU.uunet, lgr%cohoes@net.UU.uunet, mth%cohoes@net.UU.uunet Subject: Re: Sun 690 and Solbourne 700/900 In-Reply-To: Your message of Tue, 25 Feb 92 10:06:57 +0000. <5408.9202251006@subnode.arrc.salf.ac.uk> Date: Tue, 25 Feb 92 13:32:13 -0500 From: m1rcd00%fed@net.UU.uunet Status: RO

> I'm currently looking into upgrading our 4/280 to either a Sun 670 or 690 or > a Solbourne 700 or 900 Series. > > Following the "Calling 690 Managers" messages I'm wondering if the Solborne > is the better product. I'd be interested in the opinions and experiences of > this mail group.

1. We have 6 Solbourne series 5 SMP servers, and 4 S4000 workstations. By and large, our users cannot tell the difference between a Solbourne and a Sun. They are fast machines, and up until Sun released the 6xx machines, the only thing to buy for a high-end SPARC server. We probably have 100 users who depend on the Solbournes for most of their work. We use them as servers for users at NCD X terminals. The S4000 workstations are probably not worth considering.

2. We are still running 4.0.3 on the Solbournes because Solbourne was over a year later than Sun in releasing the 4.1.1 code. We are just now testing our implementation of their 4.1A.1 OS. (their version of 4.1.1)

3. Despite all of Sun's reorg fever, they still are having trouble reconciling their role as a supplier to Solbourne and their role as a competitor of Solbourne. By far, the worst manifestation of this is in getting OS patches. We have had horrible problems with the lock daemon for years, and Sun is now working on releasing patch level 8 for these problems. Every time Sun puts out a new patch, it is incompatible with the previous patch, which means that *all* machines have to be converted at the same time. But the engineers working on these patches are employees of SMCC, not SunSoft. This means that SMCC makes the change, tests the code, sends it to a few "beta" sites, certifies it, releases it as an official patch, *sends it to SunSoft*, SunSoft hacks it around a while, and sits on it. Then, once irate Solbourne clients find out about the patch, they call Solbourne, Solbourne runs into a few brick walls at SunSoft, eventually gets the source code, hacks it around for a while, tests it on their machines, and a little while later sends it out to us. All in all, it can take over a month after Sun's release date for us to get a patch from Solbourne, and it's not all Solbourne's fault. I imagine that all SPARC clone vendors are going through the same thing.

4. We had very serious hardware problems with a single Solbourne 5/60x server, which crashed on a regular basis, causing no end of trouble. To Solbourne's credit, they eventually replaced the *entire* machine even though they weren't holding the maintenance contract. But it was a long and painful road. Most of our other Solbourne machines have not had this sort of trouble, although one of the servers has had persistent disk problems that may or may not be Solbourne's fault.

5. We have had some mysterious floating point problems on the S4000 workstations.

6. Solbourne is shipping symmetric multiprocessing in thier OS. I believe that this is a big part of the reason that Solbourne's 4.1.1 was so late... they had a big release of the SMP code between Sun's 4.0.3 and 4.1.1. We ran the Solbournes for almost a year under asymmetric MP, and had some very weird problems on occasion that do not seem dissimilar to Sun's problems with the 690. SMP works much better. Sun is only shipping ASMP right now. They appear to have bagged SMP in 4.1.1 since the problem was solved in SVR4, but Solbourne has SMP in 4.1.1 today. But this means that you will probably be running 4.1.2, and ASMP, on your 690 until you can "upgrade" to SVR4, which I cannot imagine us doing until at least the middle of 1993.

7. Solbourne can put 8 40 MHz processors in the 900 server, twice as much as Sun in the 690. But processors and especially memory are pretty expensive on a Solbourne. Solbourne, I think, can also run two VME busses on a 900 server, for double the throughput in mega-IO configurations. If you need a 250 MIPS server with 50 GB of disk, the 900 may still be the only way to go. One can only imagine what sorts of Viking-based configurations Sun has on the way for late '92 and early '93, but then I can't imagine that TI would be beneath selling Viking chips to Solbourne. Whether you'll be able to put them in the 900 is another question.

8. You pays your money and you takes your chances.

Good Luck, Bob Drzyzgula, Federal Reserve Board rcd@fed.frb.gov

>From dworkin@com.rootgroup.merlin Wed Feb 26 03:40:28 1992 Received: from nsf.ac.uk by arrc.salf.ac.uk; Wed, 26 Feb 92 03:40:25 GMT Received: from csn.org by sun2.nsfnet-relay.ac.uk with SMTP inbound id <8298-0@sun2.nsfnet-relay.ac.uk>; Wed, 26 Feb 1992 03:01:02 +0000 Received: from merlin (rootgroup.com) by csn.org with SMTP id AA16333 (5.65c/IDA-1.4.4 for <djw@advanced-robotics-research-centre.salford.ac.uk>); Tue, 25 Feb 1992 20:00:42 -0700 Received: from gandalf (gandalf.rootgroup.com) by merlin with SMTP id AA18065 (5.65c/IDA-1.4.4 for <djw@advanced-robotics-research-centre.salford.ac.uk>); Tue, 25 Feb 1992 20:00:38 -0700 Received: by gandalf id AA07373 (5.65c/IDA-1.4.4 for djw@advanced-robotics-research-centre.salford.ac.uk); Tue, 25 Feb 1992 19:59:36 -0700 Date: Tue, 25 Feb 1992 19:59:36 -0700 From: Dieter Muller <dworkin@com.rootgroup.merlin> Message-Id: <199202260259.AA07373@gandalf> To: djw@uk.ac.salford.advanced-robotics-research-centre In-Reply-To: David Wheable's message of Tue, 25 Feb 92 10:06:57 GMT <5408.9202251006@subnode.arrc.salf.ac.uk> Subject: Re: Sun 690 and Solbourne 700/900 Status: RO

: Following the "Calling 690 Managers" messages I'm wondering if the Solborne : is the better product. I'd be interested in the opinions and experiences of : this mail group.

Bias statement: I used to work for Solbourne.

Going by what various people have reported, Sun's MP is where Solbourne's was three years ago (January 1989, when they did their first customer ship). This means that the MP implementation basically consists of a single lock that is taken whenever you enter the kernel for anything. Translation: the kernel is still single-threaded (can only run on one processor at a time), but multiple user-level processes can execute simultaneously.

Solbourne's MP is to the point where many operations in the kernel can execute simultaneously (for instance, you can handle a network interrupt on one processor while handling a system call on another one, and initiate disk i/o on yet a third). This means that you won't bottleneck on kernel availability nearly as much when doing i/o-intensive things.

This all relates to SunOS 4.1* and OS/MP 4.1*. SVR4 is a vastly different story (when it becomes truly available).

Dworkin Nothing is easy to solve if you try hard enough. dworkin@rootgroup.com DoD #2525 Flamer's Hotline: (303) 776-4649

>From billy@edu.unt.acs.sol Thu Feb 27 02:27:14 1992 Received: from nsf.ac.uk by arrc.salf.ac.uk; Thu, 27 Feb 92 02:27:12 GMT Received: from vax.nsfnet-relay.ac.uk by sun.nsfnet-relay.ac.uk with Internet SMTP id <g.09833-0@sun.nsfnet-relay.ac.uk>; Wed, 26 Feb 1992 06:15:29 +0000 Received: from sol.acs.unt.edu by vax.NSFnet-Relay.AC.UK via NSFnet with SMTP id aa11779; 25 Feb 92 18:28 GMT Received: by sol.acs.unt.edu id AA26428 (5.65c/IDA-1.4.4 for djw@advanced-robotics-research-centre.salford.ac.uk); Tue, 25 Feb 1992 11:24:41 -0600 Date: Tue, 25 Feb 1992 11:24:41 -0600 From: Billy Barron - VAX/UNIX Systems Manager <billy@edu.unt.acs.sol> Message-Id: <199202251724.AA26428@sol.acs.unt.edu> To: djw@uk.ac.salford.advanced-robotics-research-centre Subject: Re: Sun 690 and Solbourne 700/900 Newsgroups: info.sun-managers References: <5408.9202251006@subnode.arrc.salf.ac.uk> Status: R

In info.sun-managers you write:

>I'm currently looking into upgrading our 4/280 to either a Sun 670 or 690 or >a Solbourne 700 or 900 Series.

>Following the "Calling 690 Managers" messages I'm wondering if the Solborne >is the better product. I'd be interested in the opinions and experiences of >this mail group.

As both a Solbourne and Sun user I think I can comment. I've used both Solbourne workstation and 900 Series. On Sun, I've only done workstations and a 330 server. Points:

1. Support - Sun's support is awful. Solbourne is much better. Quicker response. On top of that, Solbourne patches (via FTP SOLBOURNE.SOLBOURNE.COM) almost always work. I've been trashed by numerous Sun patches before (like LPD).

2. Performance - the Solbourne is SMP. Sun only has MP and the word I hear from 690 managers is that the 690 on I/O bound systems perform better with less CPUs.

3. A full-loaded 700 and 900 will outperform a 690 easily. Note: Solbourne is planning to release a new line of processors during the summer which will more than double the performance of their high end systems.

4. Solbourne has been upgrade and cheaper upgrade paths.

5. Solbourne's X stuff is incredible. Puts Sun to shame. SWM has a virtual desktop capability. You can select Motif or Open Look on the fly.

As far as I'm concerned, it is a no brainer unless Sun makes you an incredible deal. Solbourne is much better all the way around. We originally were going to buy a Sun machine and then we talked to Solbourne and got a 900 instead (over a year ago). To this day, I can say it was the correct decision. :-)

Also, for me, another factor is that the Dallas Solbourne Rep Tom Hart is an excellent and honest salesman whereas our Sun rep will lie his ass off to make a sale.

Billy Barron VAX/Unix Systems Manager - and a more than satisfied Solbourne customer University of North Texas

>From matt@com.xerox.wbst845e Wed Feb 26 16:14:08 1992 Received: from nsf.ac.uk by arrc.salf.ac.uk; Wed, 26 Feb 92 16:14:05 GMT Received: from alpha.Xerox.com by sun2.nsfnet-relay.ac.uk with SMTP inbound id <18258-0@sun2.nsfnet-relay.ac.uk>; Wed, 26 Feb 1992 14:18:00 +0000 Received: from foundation.wbst845e.xerox.com ([13.2.49.99]) by alpha.xerox.com with SMTP id <11843>; Wed, 26 Feb 1992 06:10:52 PST Received: from voyager (voyager.wbst845e.xerox.com) by foundation.wbst845e.xerox.com (4.1/SMI-4.1) id AA06652; Wed, 26 Feb 92 09:10:44 EST Date: Wed, 26 Feb 1992 06:10:44 PST From: matt@com.xerox.wbst845e (Matt Goheen) Message-Id: <9202261410.AA06652@foundation.wbst845e.xerox.com> To: djw@uk.ac.salford.advanced-robotics-research-centre Subject: Re: Sun 690 and Solbourne 700/900 Sender: matt@com.xerox.wbst845e Status: RO

We have a Solbourne 800 series (a floor mount cabinet -- which they don't make any more) with three processors. In the beginning there was a LOT of trouble (SMD controller problems mostly, but MANY other) but Solbourne appears to have everything working pretty well at this point -- our server has been up 46 days! Solbourne has spent time getting the multiprocessing stuff working, and I think they are far ahead of Sun in this reguard. The Series5 processor appears slightly faster than a SPARCstation 2/IPX class processor and their serial ports and other I/O devices seem to work pretty well. The Series5e processors should be pretty nice, and I've heard that they just announced some 80 or 100 MIP superscaler thingy.

The Sun 600 series doesn't seem very impressive to me. The processors are not that fast and it looks like a stopgap planned obsolescence product to me. Then again -- I haven't really seen the thing in operation so I can't speak from experience...

- Matt Goheen

>From oran%com.amdahl.spg%com.amdahl.ccc.juts%com.amdahl.uts.amdahl@com.amdahl.charon Thu Feb 27 09:30:27 1992 Received: from nsf.ac.uk by arrc.salf.ac.uk; Thu, 27 Feb 92 09:30:07 GMT Received: from vax.nsfnet-relay.ac.uk by sun.nsfnet-relay.ac.uk with Internet SMTP id <sg.05209-1@sun.nsfnet-relay.ac.uk>; Thu, 27 Feb 1992 09:20:18 +0000 Received: from charon.amdahl.com by vax.NSFnet-Relay.AC.UK via NSFnet with SMTP id aa00595; 27 Feb 92 7:17 GMT Received: from amdahl.uts.amdahl.com by charon.amdahl.com (4.0/SMI-4.1/DNS) id AA06088; Wed, 26 Feb 92 15:44:55 PST Received: by amdahl.uts.amdahl.com (/\../\ Smail3.1.14.4 #14.9) id <m0lJYKt-0000Y8C@amdahl.uts.amdahl.com>; Wed, 26 Feb 92 15:46 PST Received: by juts.ccc.amdahl.com (/\../\ Smail3.1.14.4 #14.5) id <m0lJYJr-0000C3C@juts.ccc.amdahl.com>; Wed, 26 Feb 92 15:45 PST Received: by spg.amdahl.com (4.1/SMI-4.1) id AA21615; Wed, 26 Feb 92 15:52:22 PST Date: Wed, 26 Feb 92 15:52:22 PST From: oran@com.amdahl.spg (Oran Davis) Message-Id: <9202262352.AA21615@spg.amdahl.com> To: djw@uk.ac.salford.advanced-robotics-research-centre Subject: Sun 690 and Solbourne 700/900 Sender: oran <oran%com.amdahl.spg%com.amdahl.ccc.juts%com.amdahl.uts.amdahl@com.amdahl.charon> Status: R

Hi David,

We are using Solbourne 902/5E-128 to serve about 40 xterminal users using Motif, FrameMaker (12), Openwindows, X3270 (40), tn3270 (40), Cadre (2). We find the machine a good sun clone. The CPU performance has been adequet. We use 2.7G drives with 9MB/s IPI transfer rate which is unbeatable. Their SCSI drives run at 5MB/s which is also very satisfactory.

All in all not much problem, we are buying two more 902/5E-256.

We also have about 30 Sun SSII/GX and are happy with those boxes.

Solbourne's tech support (software and hardware) is not as good as sun's.

Please send me your summary.

>- Oran

>From era%edu.ucar.scd.niwot@edu.ucar.ncar Thu Feb 27 09:34:28 1992 Received: from nsf.ac.uk by arrc.salf.ac.uk; Thu, 27 Feb 92 09:34:15 GMT Received: from vax.nsfnet-relay.ac.uk by sun2.nsfnet-relay.ac.uk with SMTP inbound id <1045-29@sun2.nsfnet-relay.ac.uk>; Thu, 27 Feb 1992 07:21:29 +0000 Received: from ncar.ucar.edu by vax.NSFnet-Relay.AC.UK via NSFnet with SMTP id aa09371; 26 Feb 92 16:42 GMT Received: from niwot.scd.ucar.edu by ncar.ucar.EDU (5.65/ NCAR Central Post Office 04/10/90) id AA17948; Wed, 26 Feb 92 09:42:20 MST From: era@edu.ucar.scd.niwot (Ed Arnold) Received: from era.scd.ucar.edu by niwot.scd.ucar.EDU (5.65/ NCAR Mail Server 04/10/90) id AA29734; Wed, 26 Feb 92 09:42:17 MST Date: Wed, 26 Feb 92 09:42:13 MST Message-Id: <9202261642.AA02359@era.scd.ucar.edu> Received: by era.scd.ucar.edu (5.65/ NCAR Mail Client 04/19/90) id AA02359; Wed, 26 Feb 92 09:42:13 MST Reply-To: era@edu.ucar.ncar (Ed Arnold) To: djw@uk.ac.salford.advanced-robotics-research-centre Subject: Re: Sun 690 and Solbourne 700/900 Sender: era <era%edu.ucar.scd.niwot@edu.ucar.ncar> Status: R

}I'm currently looking into upgrading our 4/280 to either a Sun 670 or 690 or }a Solbourne 700 or 900 Series.

}Following the "Calling 690 Managers" messages I'm wondering if the Solborne }is the better product. I'd be interested in the opinions and experiences of }this mail group.

We had both a 6xx 2-processor and a Solbourne 2-processor in here for benching.

The conclusion: on about half our tests, the Solbourne was better. On one test, however (cat-ting stuff to 10 terminals simultaneously) the Solbourne fell flat when switched from 1 processor to 2 processors; it nice-d some of the cats so high that of ten of them, only about 4 would run at reasonable rates, unless one manually intervened and re-niced the other six processes. This seems to point to a significant deficiency in Solbourne's scheduler.

The bottom line: which is "better" depends on what sorts of things you are going to be running on the machine. ---------- Ed Arnold * NCAR * POB 3000, Boulder, CO 80307-3000 * 303-497-1253(voice) 303-497-1137(fax) * era@ncar.ucar.edu [128.117.64.4] * era@ncario.BITNET

>From miker@com.sbcoc Thu Feb 27 22:44:12 1992 Received: from nsf.ac.uk by arrc.salf.ac.uk; Thu, 27 Feb 92 22:44:09 GMT Received: from uu.psi.com by sun2.nsfnet-relay.ac.uk with SMTP inbound id <5699-0@sun2.nsfnet-relay.ac.uk>; Thu, 27 Feb 1992 21:56:30 +0000 Received: by uu.psi.com (5.65b/4.1.011392-PSI/PSINet) id AA11973; Wed, 26 Feb 92 10:12:05 -0500 Received: from trinity.sbcoc.com by sbcoc.com (4.1/SMI-4.1) id AA08645; Wed, 26 Feb 92 09:06:18 CST Received: by trinity.sbcoc.com (4.1/SMI-4.1) id AA10618; Wed, 26 Feb 92 09:06:16 CST Date: Wed, 26 Feb 92 09:06:16 CST From: Mike Raffety <miker@com.sbcoc> X-Organization: SBC/OC Services, L.P. Message-Id: <9202261506.AA10618@trinity.sbcoc.com> To: djw@uk.ac.salford.advanced-robotics-research-centre Subject: Re: Sun 690 and Solbourne 700/900 Status: R

Personally, my big problem with a Solbourne is that you're always about six months behind on OS releases AND bug fixes. IMHO.

>From mstgil@edu.unt.acs.sol Thu Feb 27 23:09:29 1992 Received: from nsf.ac.uk by arrc.salf.ac.uk; Thu, 27 Feb 92 23:09:16 GMT Received: from vax.nsfnet-relay.ac.uk by sun.nsfnet-relay.ac.uk with Internet SMTP id <g.20453-0@sun.nsfnet-relay.ac.uk>; Thu, 27 Feb 1992 12:03:25 +0000 Received: from sol.acs.unt.edu by vax.NSFnet-Relay.AC.UK via NSFnet with SMTP id aa14702; 26 Feb 92 23:28 GMT Received: by sol.acs.unt.edu id AA07345 (5.65c/IDA-1.4.4 for djw@advanced-robotics-research-centre.salford.ac.uk); Wed, 26 Feb 1992 15:18:27 -0600 From: "Marc Ph. A. J. St.-Gil - Solbourne Administrator" <mstgil@edu.unt.acs.sol> Message-Id: <199202262118.AA07345@sol.acs.unt.edu> Subject: Re: Sun 690 and Solbourne 700/900 To: djw@uk.ac.salford.advanced-robotics-research-centre Date: Wed, 26 Feb 92 15:18:27 CST Reply-To: mstgil@edu.unt.acs.sol X-Organization: University of North Texas - Academic Computing Services X-Mailer: ELM [version 2.3 PL11] Status: R

In info.sun-managers you write: >I'm currently looking into upgrading our 4/280 to either a Sun 670 or 690 or >a Solbourne 700 or 900 Series.

>Following the "Calling 690 Managers" messages I'm wondering if the Solborne >is the better product. I'd be interested in the opinions and experiences of >this mail group.

>Thanks,

>David Wheable, >National Advanced Robotics Research Centre, >Salford.

We don't have any Sun 6XX series boxes, but we do have SS2s, 490s, etc. I've been managing a Solbourne 5E/900 series server as the sole general access UNIX server for our University. We currently have about 400 users, 6 GBs of disk (2 IPI & 4 SCSI), 192 MBs of ECC RAM, and 4 processors. We've had 1 processor board go south (before we bought the 3rd and 4th) and we could've run the system on just 1 cpu, but decided not to. Solbourne Tech support has generally been much more responsive than Sun's. We have support contracts with both. With the track record of this machine it would be very hard to convice me to buy Sun system. Also, from what I've heard on the net and in the trade rags, the Solbourne systems outperform the Sun systems when you us a benchmark that's not tailored to either system. Sun's benchmarks are somewhat misleading in that they are generated with their new un-bundled compiler against Solbourne's port of Sun's old bundled compiler. I can hardly wait for the new series 6 procesors to come out and really smoke those Sun systems. There are already #defines in the kernel configuration files and c source to use the series 6 hardware, so I think they'll be out RSN. Besides, the Suns aren't true SMP systems yet, and Solbourne's been running SMP systems for 2+ years (I think) now. Also, I like the way Solbourne sends you MIT X11 compatable windowing with the OS instead of proprietary SunView, although SunView does have some nice things. I believe Solbourne is expected to release an OpenWindows product RSN too. I have it from a very reliable source that you can simply install the Open Windows 3 release on a Solbourne and it'll run just fine, only I don't think Sun's legal eagles would appreciate it.

Cheers, Marc -- Marc St.-Gil, UNIX Systems Administrator mstgil@sol.acs.unt.edu University of North Texas 817/565-2324 mstgil@vaxb.acs.unt.edu Academic Computing Services DISCLAIMER: My employers had no idea I was PO Box 13495, Denton TX, 76203 going to say that.

>From poffen@com.SLB.ATE.sj Fri Feb 28 11:36:29 1992 Received: from nsf.ac.uk by arrc.salf.ac.uk; Fri, 28 Feb 92 11:36:25 GMT Received: from ASC.SLB.com by sun2.nsfnet-relay.ac.uk with SMTP inbound id <21315-0@sun2.nsfnet-relay.ac.uk>; Fri, 28 Feb 1992 05:37:53 +0000 Received: from sjsca4.psi by ASC.SLB.COM (4.1/ASC Mailhost 3.12) id AA10322; Wed, 26 Feb 92 10:31:47 CST Apparently-To: "advanced-robotics-research-centre.salford.ac.uk::djw" X-Vms-To: advanced-robotics-research-centre.salford.ac.uk::djw Received: from rushmore.sj.ate.slb.com by eris.sj.ate.slb.com (5.59SLBATE/SLB-SERVER-1.16) id AA01216; Wed, 26 Feb 92 08:30:34 PST Date: Wed, 26 Feb 92 08:30:34 PST From: poffen@com.slb.ate.sj (Russ Poffenberger) Message-Id: <9202261630.AA01216@eris.sj.ate.slb.com> Received: by rushmore.SJ.ATE.SLB.COM (4.1/DUMB-1.0) id AA19140; Wed, 26 Feb 92 08:30:21 PST To: djw@uk.ac.salford.advanced-robotics-research-centre Subject: Re: Sun 690 and Solbourne 700/900 Sender: poffen@com.SLB.ATE.sj Status: R

> >I'm currently looking into upgrading our 4/280 to either a Sun 670 or 690 or >a Solbourne 700 or 900 Series. > >Following the "Calling 690 Managers" messages I'm wondering if the Solborne >is the better product. I'd be interested in the opinions and experiences of >this mail group. >

Are you looking for a compute server or file server? If you are looking strictly for a dedicated file server, I would look closely at the Auspex line. For file service, they beat the pants off Sun and Solbourne (IMHO) in both performance, service, and reliability. They also offer (included with the system) features like file striping and concatenation, and disk mirroring.

DISCLAIMER:

I am not am employee, nor otherwise affiliated with Auspex. I am just a satisified customer who has had an Auspex (one of the first ones) for going on 3 years now, and it continues to perform flawlessly.

Russ Poffenberger DOMAIN: poffen@sj.ate.slb.com Schlumberger Technologies UUCP: {uunet,decwrl,amdahl}!sjsca4!poffen 1601 Technology Drive CIS: 72401,276 San Jose, Ca. 95110 Voice: (408)437-5254 FAX: (408)437-5246

>From sun-managers-request@uk.ac.nsfnet-relay Mon Mar 2 19:03:57 1992 Received: from nsf.ac.uk by arrc.salf.ac.uk; Mon, 2 Mar 92 19:03:46 GMT Received: from vax.nsfnet-relay.ac.uk by sun.nsfnet-relay.ac.uk with Internet SMTP id <g.00529-9@sun.nsfnet-relay.ac.uk>; Mon, 2 Mar 1992 08:00:56 +0000 Received: from NSFnet-Relay.AC.UK by vax.NSFnet-Relay.AC.UK via List-Channel id an00699; 2 Mar 92 7:16 GMT Received: from sun2.nsfnet-relay.ac.uk by vax.NSFnet-Relay.AC.UK via NSFnet with SMTP id as00516; 2 Mar 92 7:05 GMT Received: from delta.eecs.nwu.edu by sun2.nsfnet-relay.ac.uk with SMTP inbound id <8372-0@sun2.nsfnet-relay.ac.uk>; Mon, 2 Mar 1992 03:32:59 +0000 Received: by delta.eecs.nwu.edu id AA28497 (5.65c/IDA-1.4.4 for sun-managers-outbound); Thu, 27 Feb 1992 02:10:56 -0600 Original-Sender: sun-managers-relay@edu.nwu.eecs Received: from rice.edu by delta.eecs.nwu.edu with SMTP id AA17279 (5.65c/IDA-1.4.4 for <sun-managers@eecs.nwu.edu>); Thu, 27 Feb 1992 02:10:48 -0600 Received: from rutgers.edu by rice.edu (AA00973); Thu, 27 Feb 92 02:10:05 CST Received: from aramis.rutgers.edu by rutgers.edu (5.59/SMI4.0/RU1.4/3.08) id AA26384; Thu, 27 Feb 92 03:10:34 EST Received: by dartagnan.rutgers.edu (5.59/SMI4.0/RU1.4/3.08) id AA01919; Thu, 27 Feb 92 00:15:11 EST To: ru-mlists-sun-managers@edu.rutgers Path: dartagnan.rutgers.edu!hedrick From: hedrick@edu.RUTGERS.dartagnan (Charles Hedrick) Newsgroups: ru.mlists.sun-managers Subject: Re: Sun 690MP (Credit where Credit is due) Message-Id: <Feb.27.00.15.08.1992.1900@dartagnan.rutgers.edu> Date: 27 Feb 92 05:15:09 GMT References: <9202260151.AA24692@cscgpo> Organization: Rutgers Univ., New Brunswick, N.J. Lines: 63 Sender: sun-managers-request@uk.ac.nsfnet-relay Status: R

I don't understand why people are so surprised. I thought Sun's description of the 670/690 and 4.1.2 MP support had been fairly clear. Maybe they didn't explain the consequences clearly enough, but I'm pretty sure that nobody at Sun is surprised. I'm basing my description here purely on published descriptions of 4.1.2 MP. I haven't yet had a chance to look at the kernel source, and we don't yet have enough experience with our MP systems. (We going to get some soon enough -- we're doing a flash cutover of our main campus Unix system from a Pyramid to a Sun 670 with 4 processors this weekend. This will be our first significant multiuser experience with a 670. The managers of this system are very courageous. I'm moving work onto mine in several stages, so that I can evaluate the effects.)

4.1.2 is not a full multiprocessor kernel. To provide full support for multiple processors, you have to add code to allow processors to lock kernel data structures, so that two different processors don't try to modify the same structure at the same time, etc. This is certainly possible, and many vendors have done it. But it takes time, and first releases tend to be buggy. So Sun has released an interim kernel that takes a simpler approach: it only allows one processor to be in the kernel at a time. If a process does a system call and another process is already executing kernel code, the process waits. During this wait, it can do nothing useful. (The processor probably can't switch to another process, because that would require running the scheduler, which would mean executing kernel code.)

I don't yet have enough experience to know all the consequences, but surely one obvious consequence is that the "sys" column in vmstat can never go above 25%. It's quite common for Unix systems to spend 50% of their time in the system, and under certain loads even more. A workload like that isn't going to do well on a 670 or 690 under 4.1.2. At least not if the workload stays unchanged. Of course if you move from a 490 to a 670 without changing the workload, you'll have a very lightly loaded machine. You now have 4 processors each of which is roughly the same speed as your original system. So in theory if you were spending 75% of your time in the system, you'll now be under 20% (and you'll have 75% idle time). If you use all that extra capacity to run pure CPU-bound processes, you'll be OK. But if you try to put on more work of the same kind (i.e. work that is going to be mostly doing I/O or other system activity), you will surely lose.

It's easy to believe that as system time approaches 25%, the system might go into thrashing, though I certainly can't verify that yet. One would hope that Sun would have done tests or at least simulations, and thus that they might understand the consequences in detail. But one could easily imagine that with some job mixes you'd do better by disabling processors.

I think some installations will do quite well with the 4 processor systems. You'll be fine if you are during CPU-bound things, or even if you have some I/O-bound tasks but can come up with enough CPU-bound stuff to use the rest of the compute capacity. Until the community develops more experience, or Sun reports the results of their tests in detail, about all I can suggest is that you look carefully at the "sys" column in vmstat. Both before moving to see if you've job a workload that is likely to be at risk, and after moving to see how close you are to trouble. (E.g. the Pyramid that they're about to migrate has an average of 31% user, 45% system, and 22% idle. At this instant, it's 67% system. This system appears to be at risk.)

While there are probably things Sun could do to help us manage these systems, I do feel that I was properly warned. A fully symmetric kernel will not be available until System V release 4.

----- End Included Message -----



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:38 CDT