SUMMARY: Server Machine vs. Desktop Workstation

From: Michael Steinbrunn (steinbr@informatik.rwth-aachen.de)
Date: Mon Oct 12 1992 - 19:16:08 CDT


The original question was:

>
> I'm currently assessing configuration alternatives for a small workstation
> cluster of about 10-20 machines total on a dedicated subnet.
> I've got an offer from Sun that consists
> of a SPARC 670MP Mod. 120 as server and a mixture of SS10/30, IPX and IPC
> workstations as (dataless) clients.
> Considering the cost of a 670MP, I wonder whether not some of the SS10 or IPX
> were capable of acting as servers for the cluster. I understand that the 670MP
> can support VME as well as S-Bus cards, and, thus, faster disks.
> But the SS10 now comes with a fast SCSI-2-Bus and the disks (about 2-4 1.3GB
> disks) attached to two or three SS10(-server) should yield at least the same
> performance as the single 670MP with lesser cost, even if two or three
> Prestoserve cards have to be purchased. In addition, one might even get a more
> reliable setup, because the failure of one server doesn't necessarily cause the
> entire cluster to go down.
>

Virtually all the responses I got recommended not to buy the 670MP, unless we
needed the VME bus for other reasons. The advice varied on the type of
(desktop) servers we should use: it ranges from SS1+ to SS10/54.
Apart from that, the vote was almost unanimously that (at least) a SS10 is
very well suited for our purposes. The performance of IPI disks would not
justify the extra cost--fast SCSI disks with PrestoServe are just as good.
  Another point that favours desktop workstations is lower maintenance cost
compared to the 670. Although the administration of a multi-server environment
is more complex, the overall cost/performance ratio should be much better, even
if multiple servers do not enhance reliability. One answer suggested the use of
an Auspex NS3000 as NFS server.
  Although desktop workstations do not offer the option of an NFS coprocessor
like the NC400, I think that a SS10 will give us all the performance we are
likely to need, especially as we do not plan to use diskless clients. In
addition, several responses also suggested not to run any user jobs on the
server machines to get the best possible NFS performance.

Anybody who would like the complete list of responses, please mail me.

Thanks to all who responded:
  Claus Assmann <ca@idefix.informatik.uni-kiel.dbp.de>
  adam%bwnmr4@harvard.harvard.edu (Adam Shostack)
  Gustavo Vegas <gustavo@davinci.concordia.ca>
  John DiMarco <jdd@db.toronto.edu>
  Jacques.Beigbeder@ens.ens.fr (Jacques Beigbeder)
  tommy@hogpa.ho.att.com (Thomas L Reingold +1 908 949 5706)
  higgins@math.niu.edu
  poffen@sj.ate.slb.com (Russ Poffenberger)
  rick%pgt1@Princeton.EDU (Rick Mott)
  "Jay Plett" <jay@Princeton.EDU>
  rfinch@water.ca.gov (Ralph Finch)
  scs@lokkur.dexter.mi.us (Steve Simmons)
  srm@shasta.gvg.tek.com (Steve Maraglia)
  Mariela Esser-Jaime <esser@parc.xerox.com>
  vasey@issi.com
  Peter Gray <pdg@cs.uow.edu.au>
  guyj@apollo.techpac.oz.au (Guy Jones)
  Mike Raffety <miker@sbcoc.com>
  dasun!wdceng!muir@uunet.UU.NET (Scott Muir (Muir) x6764)
  Mike Garrett <mag@jodrell-bank.manchester.ac.uk>
  fallan@awadi.com.AU (Frank Allan (Network Manager))
  celita@taux01.nsc.com
  jba@ruc.dk
  pjw@math30.sma.usna.navy.MIL (Peter J. Welcher -- math FACULTY <pjw@math30.sma.usna.navy.MIL>)
  JanBerger.Henriksen@ii.uib.no
  Christian Lawrence <cal@soac.bellcore.com>
  Michael_P._Kirby.roch803@xerox.com
  "Adam W. Feigin" <feigin@iis.ethz.ch>
  fergus@lincoln.gpsemi.com
  Bob Hendley <R.J.Hendley@computer-science.birmingham.ac.uk>

--
Michael Steinbrunn, Lehrstuhl f|r Informatik III, LuFG Prof. Kemper
Rheinisch-Westfdlische Technische Hochschule Aachen
Ahornstra_e 55, W-5100 Aachen, Deutschland
Tel.: (+49-241) 80-21340, Fax: (+49-241) 80-21349
E-Mail: steinbrunn@informatik.rwth-aachen.de



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:51 CDT