SUMMARY: How many X-terminals can a workstation serve?

From: Franz Haberhauer (Haberhauer@informatik.uni-stuttgart.de)
Date: Fri Aug 09 1991 - 12:03:07 CDT


Hi,

I had asked the Sun- and the Dec-workstation-managers how many
X-terminals one workstation can serve.

There was a consensus, that a SparcStation 2 or a DecStation 5000
can serve at least 10-20 provided that there is a lot of memory and
a lot of swap space. The more memory, the better.
A second SCSI-interface with swap space split accross two discs
improves performance too.

> How many X-terminals have you put on a workstation

                 Memory Swap # X-Terms
DecStation 3100 16 65 3 pdp@stat.ufl.edu
DecStation 5000 56 250 20 corey@cac.washington.edu
      running a tuned ULTRIX 4.1 custom kernel
      described a in a recent cross-posting to
      comp.unix.ultrix,comp.sys.dec,comp.unix.wizards
DECstation 5000/200 48 250 20 brad@cac.washington.edu
      next installation:
                     64 45
DECsystem 5000/200 64 500 25+ yzarn@chevron.com
      will add 64 MB memory

Mips RC3240 32 10 chris@bcl.co.nz
      limit for 32 MB
MIPS R6000 128 10+ Bernhard.Schneck@Physik.TU-Muenchen.DE

SparcStation 1 28 7 sgf@cfm.brown.edu
SparcStation 1+ 24 120 8 stripes@eng.umd.edu
      24 MB are not enough for 10
SPARCstation 2 64 125 15 Steve_Lammert@transarc.com
SPARCstation 2 64 10 steve@math.tamu.edu
      could probably handle 15-20

stripes@eng.umd.edu pointed out the problem of several users working
on the same machine:

> However
> if you actiavly teach a class things get _real_ slow (normally most people
> do things at diffrent times, letting others take advantage of the idle CPU,
> in a class after you say "now type cc -O4 foo.c" the load hits 8 on all 3
> CPU servers as 20 people start a compile).

> It really depends what they do. Here everything is _normally_ fine, but
> when people login things get a bit slow (while the system starts all their
> xterms and the window manager), this is mostly due to the actiave use of
> large amounts of RAM. (we have placed short sleeps in the default .xsession
> to combat this).

> Be sure to educiate the users: what they do will effect other people in the
> room immediatly. It is more like using a mainframe with a bitmapped display
> then using a workstation

Neil Gorsuch <neil@cpd.com> proposed a special solution:

> I would suggest looking at using the new Sun ELC (or whatever the
> faster SLC is called) and Graphon serial based X terminals. Both the
> ELC's and Graphon X terminals are very inexpensive. You can hook the
> Graphons at 57K baud through our SCSI based serial box. The Graphons
> have excellent data compression algorithms, and for text type
> operations such as editing and compiling, they're about as fast as
> ethernet to the eye. The overhead for SCSI transactions is much lower
> than ethernet overhead also.

Bernhard.Schneck@Physik.TU-Muenchen.DE told me, that they prefer
low-end workstation solutions because of reliability.

Many thanks to all who have replied:

birger@vest.sdata.no ( Birger Wathne)
Phil Padgett <pdp@stat.ufl.edu>
Charles <mcgrew@porthos.rutgers.edu>
Bernhard.Schneck@Physik.TU-Muenchen.DE
Steve_Lammert@transarc.com
Corey Satten <corey@cac.washington.edu>
sgf@cfm.brown.edu
Brad Greer <brad@cac.washington.edu>
steve@math.tamu.edu (Steve Johnson)
Neil Gorsuch <neil@cpd.com>
"Stripes (or Josh, if you must)" <stripes@eng.umd.edu>
Chris Mackerell <chris@bcl.co.nz>

----------------------------------------------------------------------
Franz Haberhauer haberhauer@informatik.uni-stuttgart.de
Computer Lab Manager
Institute for Parallel and Distributed High-Performance-Systems (IPVR)
University of Stuttgart
Breitwiesenstr. 20-22
D-7000 Stuttgart 80 Phone: +49-711-7816-351
Germany FAX: +49-711-7816-424
----------------------------------------------------------------------



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:20 CDT