Sat, 5 Sep 1992 02:29:15 +0200
|
>SHOW BITEARN command
> Compiled Regular
>CPU sec. .15 .62
>Paging 61 16
>WS (observed maximum IND USER) 225 174
LSVSHOW was not EXECLOADed by the version used for this test, otherwise
the difference in CPU time would be irrelevant due to not having much to
do for this command; here what you are seeing is the time to load a big
REXX source into memory one line at a time vs the larger LRECL of
compiled files. The interpreter wins on paging and WS because the
compiled version is usually about 5 times larger. I don't recommend
compiling LSVSHOW on storage constrained systems unless you use the
command very often.
>NODESGEN command
>
>CPU 39.37 113.62
>Paging 127 82
>WS 377 326
That one is now in PREXX, although some of the post-processing (PEERS
DISTSUM, REFRESH, etc) is still in REXX and amounts to half of the total
CPU time.
-------------------------------------------------------------------------
Date: Sat, 5 Sep 1992 02:03:18 +0000
From: Revised List Processor (1.7d) <[log in to unmask]>
Subject: Output of your job "ERIC"
To: [log in to unmask]
> nodesgen
Nodes tables generation process started - be patient...
(...)
Nodes tables generation completed successfully.
Summary of resource utilization
-------------------------------
CPU time: 2.463 sec Device I/O: 530
Overhead CPU: 0.325 sec Paging I/O: 0
CPU model: 3090 DASD model: 3380
-------------------------------------------------------------------------
Note: the figures cannot be compared directly because the CPU's are not
the same and the 1.7 code has more work to do due to new functions and
the data conversion required for the new format ':links' tags.
Compiling LSVBITGN won't help much with 1.7d, it is LSVPRSUM that eats
most of the 1.5 sec burnt up by REXX code. But then who cares about
fractions of seconds a few times a month?
>DIST2 job (4 local, 1 remote recipient)
>
>CPU .40 1.61
>Paging 59 7
>WS 269 216
Compiling LSVDIST2 helps a lot, because of all the nested loops. If you
are going to compile one EXEC, this is the one to choose. The second is
LSVXMAIL. Other good choices are LSVCKPRV, LSVBLC and LSVMAIL (for 1.7d).
If you compile *any* EXEC, please also compile LSVPROF. This is the EXEC
that sets the "compiled" flag you can query with the RELEASE command.
Compiling LSVPROF won't save you any time, but it's only 120 lines long
and won't cost you much memory.
Eric
|
|
|