Xref: utzoo comp.windows.x:18330 comp.dcom.lans:4333 comp.unix.wizards:20571
Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!iuvax!cica!tut.cis.ohio-state.edu!ucsd!ucsdhub!celit!dave
From: dave@fps.com (Dave Smith)
Newsgroups: comp.windows.x,comp.dcom.lans,comp.unix.wizards
Subject: Re: X terminals on an ethernet
Message-ID: <6846@celit.fps.com>
Date: 13 Feb 90 03:04:58 GMT
References: <2519@bacchus.dec.com> <1990Jan26.044522.10299@Solbourne.COM> <1990Feb7.201302.3346@sci.ccny.cuny.edu>
Sender: daemon@fps.com
Reply-To: dave@fps.com (Dave Smith)
Followup-To: comp.windows.x
Organization: FPS Computing Inc., San Diego CA
Lines: 49

In article <1990Feb7.201302.3346@sci.ccny.cuny.edu> dan@sci.ccny.cuny.edu (Dan Schlitt) writes:
>The plan is to share a stardent titan among the users.  How many X
>terminals is it reasonable to hook on the ethernet?  The question is
>not really whether the host can handle it but whether the network can
>handle it.  Three, 10, 40?
>
>Another way to phrase the question is "how does X client-server
>traffic compare with NFS traffic for diskless workstations?"  Folk
>here have some feelings about the limits of diskless workstations and
>would understand that kind of comparison.
>
>Thanks for any help you can give us.

I'd say that this would really depend on what your users are planning to
do with the terminals.  If you're planning to use them to run lots of
xterms, I'd say the overhead will be minimal.  Xterm doesn't generate
much more overhead than rlogin.  Graphics applications, well, that would
depend on what size images and how often they're transfered.  If you're
comparing X-terms to diskless workstations the comparison is image size
vs. amount of data required to generate the image, i.e. if you have to
process 20M of data to create a single 1M screen image the X terminal 
will be a big win.  If, on the other hand, you had a small data set that
would fit in the memory of a diskless workstation and the data was used
to generate a large number of pictures, the workstation would generate
less network traffic.

Our network here has over 50 Suns on it, plus about 10 larger machines.
The Suns all have local disks for swapping, but load most of their objects
across the network.  Usage of the Suns is about half as windowing terminals
(rlogin's to the main machines for us software critters) and about half as 
CAD stations, for the EE's.  About five of the Suns run X and we have four 
or five more X terminals on the network.  In addition, there is a _lot_ of 
NFS traffic between the larger machines.  We're really pushing the Ethernet 
(it's still a single net, sigh...) but things run OK.

So, as far as the number of X terminals a network can handle, I'd say it
really depends on the type of application.  For just xterms, I'd say
30 or 40 could be put on a net without any performance problems, maybe more.
(Ours runs fine and we have a lot of NFS in addition to rlogin/xterm stuff).
For heavy duty graphics stuff you'd have to figure what the average load
per X terminal would be and then divide that into the network speed (be
real pessimistic, assume 250K/sec or so.  I've never seen anyone bitch 
because things run faster than you told them they would :-) ).

--
David L. Smith
FPS Computing, San Diego
ucsd!celerity!dave or dave@fps.com
"I'm trying to think, but nothing happens!" - Curly Howard