jffowler@gamera.cns.syr.edu (John F. Fowler) (10/04/89)
A week ago I posted a query about the feasibility of using a 4/390 for a general timesharing compute server. I received quite a few responses, most of which were requests for me to share any answers I got ..... apparently others are thinking along the same lines. Generally, of those that have looked into it, the responses are positive. There were 2 mailings from people that had tried it and were unsuccessful; here they are: >> >> Date: Wed, 27 Sep 89 13:17:03 -0400 >> To: "John F. Fowler" <uunet!gamera.cns.syr.edu!jffowler@uunet.UU.NET> >> From: jam@philabs.Philips.Com (John A. Murphy) >> Subject: Re: SUN 4/390 as Compute/Timesharing Server? >> >> My experience with a 4/280 is it would grind to a halt under those >> conditions. We have had just a few users running compute bound simulations >> and it can cause several server dropouts to clients. Response on the >> machine also slows considerably. I can't comment about a 390. >> >> Murf .. >> >> Date: Sat, 30 Sep 89 21:11:06 EDT >> To: jffowler@gamera.cns.syr.edu >> From: jdh@bu-it.BU.EDU >> Subject: re: SUN 4/390 as Compute/Timesharing Server? >> >> We don't have any 4/390s, but we have several 4/280s >> >> My experience has been that they don't fair very well as timesharing >> engines; actually, and I may be in the minority with this, the 3/280 >> seems to better handle generic light timesharing for n users ( where n >> equals about 10 to 20 users ) >> >> There's been a lot of conjecture as to this, but the best reason that >> I've heard is that the sun4 has 7 (8?) register windows and this >> affects the rate at which it can do context switches. When you have >> more than 7 or 8 active users performance is sluggish. >> >> If you're looking for just a MIPs box with light to medium timesharing >> it works pretty well ( though the 4/280 is probably somewhat less than >> the rated 10 MIPS, maybe 8 or 9 MIPs ) >> >> Jason Heirtzler >> Distributed Systems Group >> Information Technology >> Boston University There were 2 responses from enthusiastic Solbourne users. Both of them were essentially doing this in addition to other tasks with their Solbournes and were happy. They were Paul Graham <pjg@urth.acsu.buffalo.edu> and David Botticello <botticel@orion.crd.ge.com>. There were 9 responses from people that were using 4/260, 280, or 390 hardware successfully for this purpose. One of the constant themes in these messages was the need for adequate memory. Everybody that mentioned it was using a minimum of 32MB. Mark Foster of UPenn has a rule of thumb of 1MB per user, which seems to coincide with what other people are doing. Here are the respondents: Ken Rossman <ken@watsun.cc.columbia.edu> - 4/280s, 15-30 academic users per machine, very happy. Scott Schwartz <shire.cns.psu.edu> - two 4/280s, typically 10 users & some crunching. David Oliver <oliver@smectos.gang.umass.edu> - 4/260, serves 5 diskless, 4 diskfull, 20-40 timesharing users, running pretty full. Mark Bell <markb@jpl-cray.jpl.nasa.gov> - 4/280, 30-40 users plus large Informix database jobs; pleased. Charles Buchholtz <chip@pender.ee.upenn.edu> - 4/280, typically 15 users and some large number crunching (SPICE). Mark Foster, <mark@central.cis.upenn.edu>, Ira Winston <ira@central.cis.upenn.edu> - 4/280s, light timesharing with email and text formatting, 40-50 logged in, load average usually less than 1 unless dumping. Vaughan Pratt <pratt@coraki.stanford.edu> - About to go production on switch from Vax 8600 to 4/390 after lengthy testing. Craig Warren <munnari!charlie.oz.au!ccw@uunet.uu.net> - 4/260s, 4/280s, 30-40 students apiece with database work.