Path: utzoo!attcan!uunet!snorkelwacker!bloom-beacon!EXPO.LCS.MIT.EDU!keith
From: keith@EXPO.LCS.MIT.EDU (Keith Packard)
Newsgroups: comp.windows.x
Subject: Re: X/Windows Performance Query
Message-ID: <8910220252.AA09500@expo.lcs.mit.edu>
Date: 22 Oct 89 02:52:24 GMT
References: <1989Oct19.173630.20790@hellgate.utah.edu>
Sender: root@bloom-beacon.MIT.EDU
Organization: The Internet
Lines: 38


Brad Hutchings (hutchings@cs.utah.edu) writes:

>    I am trying to find out what performance I can expect from X11
> on a 386 AT-clone relative to a vanilla 68020-68030 UNIX workstation
> such as a SUN or an HP. *Any* opinions are welcome.

This question begs some more information.  It is very important for you
to describe what sort of operations you will be using X for.  Asking
for the "best" X server is like asking for the "best" workstation.

If you are interested in simple user-interface activities (solid fills,
bitblt, text painting), the answers will be much different than an
application mix which makes extensive use of complex graphics (wide lines,
filled arcs, filled polygons).

In the first case, some of the current crop of X terminals with simplistic
graphics co-processing out-perform many available workstations.  In the
second case, a machine with additional graphics compute power (Faster
general purpose CPU or possibly a more complex graphics co-processor) would
be suitable.

Beware of published performance numbers, they frequently talk about raw
graphics performance (vectors/sec, chars/sec) and neglect to add in the
overhead required to convert X requests into the low-level primitives.

The best way of getting the most performance/dollar is to benchmark the
actual applications you'll be using on a variety of platforms.  Lacking this
capability (not too many people want to port a large collection of
applications), you can use a benchmark program (like x11perf) which will
allow you to examine some individual aspects of the system which are
applicable to you needs.

A single number is never appropriate as a characteristic of system
performance.

Keith Packard
MIT X Consortium