home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
TeX 1995 July
/
TeX CD-ROM July 1995 (Disc 1)(Walnut Creek)(1995).ISO
/
tex-k
/
tex-k-archive.past
/
tex-k-archive.gz
/
tex-k-archive
/
000843_faith@cs.unc.edu_Fri Aug 12 05:46:39 1994.msg
< prev
next >
Wrap
Internet Message Format
|
1994-10-11
|
2KB
Received: from mcenroe.cs.unc.edu by cs.umb.edu with SMTP id AA19468
(5.65c/IDA-1.4.4 for <tex-k@cs.umb.edu>); Fri, 12 Aug 1994 09:47:26 -0400
Received: from proteus.cs.unc.edu by mcenroe.cs.unc.edu (8.6.9/UNC_06_21_94)
id JAA11531; Fri, 12 Aug 1994 09:46:40 -0400
Received: by proteus.cs.unc.edu (8.6.9/UNC_06_21_94)
id JAA04116; Fri, 12 Aug 1994 09:46:39 -0400
Date: Fri, 12 Aug 1994 09:46:39 -0400
From: Rik Faith <faith@cs.unc.edu>
Message-Id: <199408121346.JAA04116@proteus.cs.unc.edu>
To: tex-k@cs.umb.edu
Subject: TeX and DEC Stations
Cc: faith@cs.unc.edu
Our department has several hundred DECStation machines. These are based on
a MIPS chip, run ULTRIX, and are about 1/2 as fast as a i486/33. We also
run AFS, which tends to slow down initial file access and the stat'ing of
files. Since we use very small AFS caches on the DECStations, AFS tends to
make file accesses slower than under NFS. Since ULTRIX uses a fixed size
buffer cache, and since most of the machines only have 16MB of RAM, the
UNIX buffer cache is probably worthless from a performance standpoint.
I've compiled the distribution (lib-6.1 and src-6.0) for these machines
using gcc 2.5.8 and the -O option. TeX is incredibly slow (about 3 to 5
times slower than our old TeX 2.98 installation). This leads to a lot of
complaints about performance. Many users refuse to use the new TeX
installation. (Using the latest kpathsea package does not help much.)
LaTeX 2e is even slower than plain TeX or LaTeX 2.09.
I'd like to recompile TeX for the DECStations, with the goal of improving
performance. I hope that someone on this list can give me some pointers on
how to improve performance. Will using MIPS C instead of gcc help? Should
I use -O2? Should I unroll loops? Omit frame pointers? Use any other
uncommon optimizations?
Thanks for any help and insight, Rik.