SUMMARY: Questions on Multi-Processor SuperSparcs

From: Robert N. Thompson (rnt@math.ias.edu)
Date: Fri Apr 09 1993 - 20:59:41 CDT


Sun Managers,

        Here are the results that I have received to date. The original post is
at the end of the message. Many thanks to all who sent information!

1) No one reported having a Model 52 yet, though some managers have them on order.

2) Only one person <ric@is.morgan.com> reported having a dual CPU setup running
Solaris 2.1, his results:

=> Several months ago I put two model 41 MBUS cards into
=> a SPARCstation 10 chassis and loaded Solaris 2.1.
=> The ROM and OS saw and used the dual CPUs without
=> any problem. I experienced no kernel panics or other nastiness.
=> Solaris 2.1 does have different performance characteristics
=> than SunOS 4.1.3 but that's a different story. More
=> interestingly, the OS was clearly multi-threaded in most
=> of the right ways: throughput of many system call benchmarks
=> (fork, exec, context-switch, system-call speed) was almost
=> exactly twice that of a single CPU machine. For people
=> who need throughput performance of multiple processes, this
=> combination ought to be a win.
=> Note that Solaris 2.2 with more performance enhancements and
=> a user-level threads package will be out in May.

3) Most people reported no hardware/software problems with dual CPU setups.
The one exception came from <hedrick@cs.rutgers.edu>:

=>The main problem we've seen is Common Lisp -- a couple of versions are
=>built by saving a core image. If you try to run the code on a system
=>where the stack is at a different location, it doesn't work.
=>Unfortunately the Viking puts the stack in a different place from the
=>SS2 and older 670 processors. So our old Common Lisp images wouldn't
=>work. Franz, Inc., gave us a fix that makes the saved images not be
=>sensitive to changes in stack location. So that problem has been
=>fixed. We've also run into a problem with an old version of Ingres.
=>It works fine on an SS2, but not an SS10. Apparently it's trying to
=>mmap a page of memory at a fixed address. Again, memory layout is
=>slightly different on the Viking, so the old address doesn't work. I
=>assume Ingres would fix this for us, but in this case we don't have
=>software support.
=>
=>Both of these problems are due to the Viking processors themselves,
=>not to having multiple processors. We haven't seen any problems due
=>to having multiple processors so far.
=>

4) Solaris 2.2 has just been announced and is supposed to significantly increase the
performance of multi-processor machines.

5) <tlr@toy.rad.msu.edu> reported running a model 41 SS10 with 2 40MHz single-cpu
modules in it under SunOS 4.1.3. But since it lacks symmetric MP it probably doesn't
buy much in the form of performance gains.

Again thanks to these Sun-Managers:

hedrick@cs.rutgers.edu
cogan@merlin.anu.edu.au
feigin@iis.ee.ethz.ch
hkatz@NUCMED.MED.NYU.EDU
eckhard@ts.go.dlr.de
ric@is.morgan.com
shandelm@jpmorgan.com
tlr@toy.rad.msu.edu
jjd@spserv.bbn.com

Bob Thompson
Sun Manager
School of Mathematics
Institute for Advanced Study
South Olden Lane
Princeton, NJ 08540
 
email: rnt@math.ias.edu
phone: (609)-734-8331
fax: (609)-924-8399

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

        We are in need of any real-life experiences with either the SS-10 Model 52
or a Model 41 with a second SuperSparc Processor. (Even the 6xx series with multiple
SS processors would be relevant)

1) Does anyone have either of these systems installed on an existing Sun network?
2) If so, have you experienced any MAJOR problems with installation, OS operation,
        compatibility problems, etc..
3) Do you have any performance figures for the dual CPU workstations/servers? ie:
        are they close to the advertised numbers?
4) Any other related information....

        We are about to purchase a high-end machine and could really use any timely
and useful information about these units before spending the $!

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



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:07:42 CDT