In article <[log in to unmask]>, Gary Sponseller
<[log in to unmask]> says:
>
>Should I just go for all 126?
>
>Or should I use the most-often-used EXECs from the results from a
>CMS EXECMAP command on the server console?
>Currently, I see only 20 heavily-used EXECs on this map...
>
>Do very short EXECs matter?
 
I can't speak for LISTSERV, but here's my experience with PSU NetNews
(usenet news for CMS, mostly large REXX execs and xedit macros):
 
Of the 8 or so major execs used in the server, I compiled the ones that
did the bulk of the processing, and saw a significant drop (see below)
in CPU time required.  In both cases, the subroutines were execloaded.
 
The client programs used 2/3 less CPU when I compiled four selected
routines (there are probably a dozen or so altogether).
 
If you compile everything, you'll need more disk-space, and my brief
experiments showed trivial gain on infrequently-used routines.  It also
required more VMSIZE for both server and client, particularly if you
reply to an article with 'MAIL' that's been compiled, as well.
 
Another observation: For large associative lookup tables (and I don't
know if LISTSERV uses these or not), search time is more linear in
compiled execs than interpreted ones.  For example, NetNews keeps a
g.msgid.<received_article> = 1 entry for each article already received.
G.msgid starts off at 0, so you can test to see if an article has already
been processed 'if g.msgid.<test_msgid> = 1 then its_a_duplicate'.  We
routinely process 8-12K articles per day.  I believe this was discussed
on REXXLIST about a year ago.
 
Specific numbers - avg CPU per article, Amdahl 5850:
 
interpreted, near startup  - .25 sec/article
interpreted, close to shutdown - .95+ sec/article
compiled, near startup - .10 sec/article
compiled, close to shutdown - .16 sec/article
 
Since the times are averages, it was taking over one second per article
by the end of the day.  Before I compiled the code, I took to rebooting
the server at lunch-time to cut the resource usage down.
 
Having said all that, I wouldn't bother compiling everything.
 
Doug
--
Doug Sewell, Tech Support, Computer Center, Youngstown State University
[log in to unmask]   [log in to unmask]   ...uunet!ysu.edu!doug
I support drug testing. I believe every public official should be given a
shot of sodium pentathol and ask "Which laws have you broken this week?".