Fri, 4 Sep 1992 18:07:23 GMT
|
In article <[log in to unmask]>, "Jonathon Simon"
<[log in to unmask]> writes:
|> What's the latest on REXX compilation of LISTSERV. Is it worth it?
|> Does it have any significant impact on performance. We are looking
|> for anything that can help drop our CPU level during prime time.
|>
|> Thanks in advance.
I have compiled it with both IBM REXX and PROREXX (from Systems Center). We
decided not to purchase either of these packages, but here are some numbers
from LISTSERV 1.6e(?) compiled with the IBM REXX compiler on a 4381-13 with
3380 DASD under SP 5:
SHOW BITEARN command
Compiled Regular
CPU sec. .15 .62
Paging 61 16
WS (observed maximum IND USER) 225 174
NODESGEN command
CPU 39.37 113.62
Paging 127 82
WS 377 326
DIST2 job (4 local, 1 remote recipient)
CPU .40 1.61
Paging 59 7
WS 269 216
The PROREXX compiler was much faster at compiling the programs, but the CPU
times were slightly higher for PROREXX (approximately 50% of the uncompiled
times) and the working set requirements (both idle and busy) were higher. I
helped Systems Center fix several bugs in the compiler, too (mostly buffer
overflow situations). The IBM compiler was SLOW, but I guess you only have
to do it once.
Though we could, of course, use any extra CPU, our machine's primary
performance problems come from paging/real storage demands. So, I would
rather use twice as much CPU and half the memory.
Eric is gradually switching LISTSERV functions to compiled HLL modules, so
the need for compiling the REXX functions is reduced with every release.
--Shane
|
|
|