laser-lovers@uw-beaver (laser-lovers) (09/22/84)
From: chris@maryland (Chris Torek)
Here is a benchmark from William Sebok, comparing TeX and troff speeds
(remailed by permission). The machine is a VAX 11/750 running 4.2BSD.
>From astrovax!wls Tue Sep 18 22:55:08 1984
Subject: TeX versus troff benchmark.
The results of a race-off between TeX and vtroff are as follows. The file
is a tex chapter of Todd Lauer's thesis translated into troff'ese by Bruce
Draine.
TeX: cpu time
tex 77 sec
dvipr (1st pass) 22 sec
verser2 (called by spooler) 43 sec
---- -------
Total 142 sec
roff
roff (tbl + eqn + vtroff) 175 sec
rvcat (called by spooler) 36 sec
---- -------
Total 211 sec
Conclusion: In this benchmark TeX uses 33% less cpu time. Whether this is
enough of a difference to make you switch is up to you.
[And a small appendage]
"roff" here is a local shell script written by Bruce Draine that
invokes tbl, eqn and vtroff and adds in some local macros.
--- Bill
--
(This page accidently left blank.)
In-Real-Life: Chris Torek, Univ of MD Comp Sci (301) 454-7690
UUCP: {seismo,allegra,brl-bmd}!umcp-cs!chris
CSNet: chris@umcp-cs ARPA: chris@marylandlaser-lovers@uw-beaver (laser-lovers) (09/24/84)
From: harrison%ucbrenoir@Berkeley (Michael Harrison) It seems tricky to find fair ways to compare TeX and troff. It seems that the source should contain tables and math since that is built in to TeX and that troff should be charged for tbl and eqn. But the inclusion of the printing time seems to complicate the comparison. I would suggest that the fairest comparison would involve using the -t option to troff to just create a file. Lastly, can anyone dream up an input which includes tables and math in a nontrivial way that scribe could do so it could be included into the comparisons?
laser-lovers@uw-beaver (laser-lovers) (09/24/84)
From: chris@maryland (Chris Torek)
Certainly it's difficult. One timing doesn't say a whole lot. The
only way you can tell for sure which one is faster for your own papers
is to try both. Of course, this is impractical for most people.
However, I don't think the postprocessing time should be discounted,
since that is a (sometimes quite large!) factor in the overall time
consumed. Again, though, this time will depend on what you are doing.
For example, the Versatec driver must read in a complete page, sort it,
and then reprocess it later to push bits onto the page. An Imagen
driver can be one pass and would be considerably more simple, since one
just downloads glyphs then invokes them at the proper position.
If you *are* using an Imagen, there is something else to think about:
the Imagen software for troff works by converting the C/A/T typesetter
codes to DVI files (TeX output), which is then reprocessed into imPRESS
(Imagen's typesetting codes). Since TeX output doesn't use this
intermediate process, the postprocessing time will naturally be less.
(Side note: catdvi's output is not quite compatible with TeX82's
output, so Imagen's ``dviimp'' program doesn't work for TeX. Instead,
we're using a temporary patch to Mike Urban's driver to make it work on
8/300s. *Sigh*)
Then again, if you actually have a C/A/T systems phototypesetter, troff
would probably be faster than TeX. . . . (Also, there is the fact that
troff is much smaller than TeX -- at least, if you're using the
standard one (ours has *BIG* internal buffers and runs around 700K).
And I doubt that anyone has TeX running on PDP-11's.)
Anyway, to get back to the point, it's not a simple issue and I just
mailed out those timings as an interesting statistic (hence the subject
line).
--
(This page accidently left blank.)
In-Real-Life: Chris Torek, Univ of MD Comp Sci (301) 454-7690
UUCP: {seismo,allegra,brl-bmd}!umcp-cs!chris
CSNet: chris@umcp-cs ARPA: chris@marylandlaser-lovers@uw-beaver (laser-lovers) (09/24/84)
From: Brian Reid <reid@Glacier> When you are comparing speeds of text formatters it is important (though difficult) to take into account the number of times that the average user has to run his job in order to get it formatted correctly. If it often comes out right the first or second time, it is less important that the processing times are longer. Could you send me a copy of the benchmark? I'd like to push it through Scribe and see what its timings are. In similar tests in the past Scribe has been slightly faster than TeX, but timings of course vary widely depending on what you are doing.
laser-lovers@uw-beaver (laser-lovers) (09/24/84)
From: chris@maryland (Chris Torek) I don't have it [the text that was used for the timings]. If you can get mail to astrovax!wls, he can probably mail you a copy. Chris