Path: utzoo!attcan!uunet!cs.utexas.edu!rutgers!apple!bloom-beacon!SUN.COM!dshr
From: dshr@SUN.COM (David Rosenthal)
Newsgroups: comp.windows.x
Subject: Re: Faster color X for Sun-3s?
Message-ID: <8904281800.AA00273@devnull.sun.com>
Date: 28 Apr 89 15:25:54 GMT
Sender: daemon@bloom-beacon.MIT.EDU
Organization: The Internet
Lines: 59

Steve Allen from UC Santa Cruz writes:

> I am anxiously awaiting the release of X11/NeWS from Sun along with everybody
> else who has a Sun and wants a faster server.  Unfortunately, I am sitting
> in front of a 4 Megabyte Sun.

X11/NeWS is faster than the MIT sample server,  but the difference is not
overwhelming for monochrome machines.  A 4M color Sun has performance problems
irrespective of the window system it is running.

> At the announcement party for the new Sun
> Sparcstations 2 weekes ago, Sun confirmed that they do not recommend running
> X11/NeWS on a 4 Megabyte Sun unless you use it *only* as a window onto other
> machines.  I asked if Sun was planning to release a version of vanilla
> X11 without NeWS which was optimized for Sun machines.  The answer was, "No."
> It was a qualified `No', but still a `no'.

X11/NeWS uses more memory than the MIT server (like most commercial X servers)
but it does so largely to improve rendering performance,  not to implement the
NeWS interpreter.  As Robin Schaufler explained in her Usenix paper,  X11/NeWS
is not implemented by adding a PostScript interpreter to a basic X server,  but
by implementing an integrated window server capable of supporting multiple
interpreters and both X and PostScript imaging models.  "Configuring out"
NeWS (or X) would result in only a small space saving - basically just the
protocol interpreter loop - and isn't really practical.  This should be
contrasted with the approach taken by vendors adding Display PostScript
to X servers - since there is little integration between the interpreters
and imaging systems configuring out DPS will save a lot of space.

Further,  I think you have mis-represented Sun's position somewhat.  I believe
that what was said was that Sun does not recommend running OpenWindows(TM)
with both the server and clients on a 4M machine.  This restriction is due
more to the resource requirements of the OpenWindows clients (the replacements
for the familiar shelltool,  cmdtool,  and so on implemented using the XView
toolkit) than to the resource requirements of the X11/NeWS server itself.  If all
you run are vanilla X clients like xterm,  the resource requirements will be
significantly less.

> It would seem that we who have
> 4 Megabyte machines will have to rely on MIT and Purdue, or go to a third
> party vendor.
> 
I have a 4M monochrome 3/75 and run NeWS,  MIT X and X11/NeWS on it (not all at
once :-).  For vanilla X applications,  the overall system performance is
currently slightly better with the MIT sample server than with X11/NeWS.  I would
strongly urge anyone with a 4M color system to add memory without regard
to the window system they run.  If they do,  they will find that X11/NeWS
is substantially faster than the MIT sample server.

The performance of an X server is largely determined by the performance
of the DDX imaging layer.  It is greatly to the credit of the server
implementation team at DEC that the "mfb" code,  while highly portable,
provides good performance.  Although optimizations are certainly possible,
it is unlikely that anyone (Sun, MIT,  or a third party) would be able to
provide a version of "mfb" that would provide radically better performance
in a monochrome 4M machine.  Everything I can think of to improve mfb's
performance costs memory,  which is where we came in....

	David.