CPU time is snowballing - possible causes?

nuck · New User Joined: 09 Dec 2005 Posts: 33

hi fellow listies,

I've got an exec that is exhibiting some rather odd behaviors (not entirely unlike the programmer that wrote it

).

The CPU times that it takes to run through the main loop keeps getting longer and longer. For example, if I run it with a loop of 'do j=1 to 100' the first ten iterations are relatively quick (under 1 second CPU), after the 10th loop it's up to over 10 seconds, after the 20th, it's ove 30 seconds per 'do j'. I ttested this by changing the loop condition to 'do j=30 to 35', and these iterations run just as quickly as the iterations 1 thru 5 in the first test. huh???
Seems to indicate that something is stealing storage somewhere, not letting it go, and slowing everything down.
Any ideas what it could be?

thanx in advance!

ofer71 · Posted: Mon Aug 20, 2007 3:02 pm

Can you post the entire loop?

O.

nuck · New User Joined: 09 Dec 2005 Posts: 33

ummmm...not really...

well, I could, but it's about 600 rows of code...I'm currently checking if it's caused by compound vars not getting re-initialized - altho that shouldn't really be causing this much trouble (?)

I suspect it's from a sort that I call (ICEMAN) where I'm reallocation the sysin, sysout, etc every time thru. That would slow it down a bit, but so much?

I'm gonna put in some tracing infos that track the cpu usage at various points in the prog, that should help me narrow it down.

I wasn't really looking for the exact solution to this prob - more of a 'what could it be?' question. I haven't had something like happen in Rexx yet, and I've been playing with rexx for damn near 20 years.

ofer71 · Posted: Mon Aug 20, 2007 4:52 pm

How about QUEUEs? Any ISPF services? ISPF table manipulation?

O.

nuck · New User Joined: 09 Dec 2005 Posts: 33

oops

looks like another classic of case PEBCAK....

I jumped the gun on this one - turns out I should've finished my testing...there is a problem with my initializations, which means my compound vars are just getting bigger and bigger and bigger - which would also explain the snowballing effect on cpu time.

sorry guys, my fault. Back to the drawing board.

Indian-Delta-10-Tango....(referring to myself)

dick scherrer · Posted: Mon Aug 20, 2007 9:45 pm

Ahhhh. . . another instance of the I-D-ten-T error

I hate when i raise those

But just who is PEBCAK - as in "another classic of case PEBCAK"?

dbzTHEdinosauer · Posted: Mon Aug 20, 2007 10:31 pm

I must be an I D 1 0 T, but I don't under stand I-D-ten_T error.

I hate it when I am one.

Like the way I am feeling now, trying to figure out PEBCAK....

dick scherrer · Posted: Mon Aug 20, 2007 11:19 pm

nuck · New User Joined: 09 Dec 2005 Posts: 33

heh...

PEBCAK= Problem Exists Between Chair And Keyboard.

phonetically I used the NATO alphabet (alpha, bravo, charlie, etc) for the Indian Delta ten Tango.

I love it when someone asks me to spell out my last name - I do it phonetically, which is kinda long, cuz my last name is 13 characters long. Usually after the fifth character they're yelling at me to stop.

agkshirsagar · Posted: Tue Aug 21, 2007 2:16 pm

Shouldn't this topic be moved to Fun stuff????

nuck · New User Joined: 09 Dec 2005 Posts: 33

PS: solved the bug in my exec. Now it takes about 10 CPU minutes from start to end to analyze 280 JCLs, including the building of cross-references and dependencies. YAAAY!!!!

dick scherrer · Posted: Tue Aug 21, 2007 10:20 pm

Good to hear that the exec now runs successfully

Ten minutes cpu time still sounds like a lot. . . That comes to just over 2 cpu seconds per jcl. Even fixed, it sounds like an intense exec.

nuck · New User Joined: 09 Dec 2005 Posts: 33

yup, most of the JCLs take less than 2 seconds, the JCLs with 50 steps tend to chew up a lot of cpu time.

my exec creates documentation for the operations boys, a listing per job and step of which files are used, with a cross reference for each DSN to all other jobs (we can then see where the dsn is created), as well as the scheduling dependencies.

so, it does have a fair bit of processing to do.

...and before the flaming starts: Yes I know docu is usually created BEFORE a program or JCL is written. However, this method guarantees that the docu matches EXACTLY to the code. And it removes the possibility of human error while writing said docu.

PS: I'd love to find a job where I write more REXX stuff....it's just too much fun.