On Dec 27, 2007, at 1:12 PM, Ted MacNEIL wrote:

I rewrote a single COBOL program myself. The run time went from about 12 hours to less than one hour. And it was a change to only one paragraph. And not likely a truly optimal change at that.

In the very early 1980's, my ex-wife was a junior COBOL programmer at a large Canadian (now out of business) retailer, in Toronto.

They had this huge batch job that took (literally) hours to run.
With the multiple time-zones in Canada, this was becoming a major problem.
So, they assigned her the job of optimising it.
They didn't know her husband was a performance analyst.

So, she brought the problem to me.
I asked her to show me the JCL and the File Definitions.
I know COBOL, but I hate debugging it.

All the files were FB-80-80!
I told her to change the FD's to BLOCK CONTAINS ZERO.
And, to maximise the blocksizes in the JCL.

She did it in two steps.
COBOL first, JCL second.

When the job ran for the first time, after both changes, it took less than 20 minutes. She was actually paged because Operations had thought it had failed because of the shortened duration even with RC=0 for all steps.

She got a raise and a promotion out of this, because it was running at a service bureau and saved real money, including disk and tape.

So, the point: not all optimimisation is code.
CPU is cheap compared to I/O, and I/O can be optimised with common sense. CPU optimisation takes analysis and an understanding of what the code is doing.
I/O just takes looking at the file definitions.

-
Ted,

Good story. I think most have had similar stories. I definitely can identify with the records contain item. Somewhere in the dark past a fellow sysprog wrote a program to find those types of *STUPID* errors in the source. It worked when source was kept in a PDS but could not touch Panvalet. There was (IS?) an option in Pan to create a PDS from Panvalet and I actually found 23 programs that did that. We got that changed in the programs (after some political grief) and the production cycle dropped by 2 hours. The programmers took the credit, another political gaff. The management turned around and said they didn't need a new CPU so 6 months go by and again we are hitting the "window" (options market doesn't open if we don't get done by the window and we get fined big time) I dive back in and find even more of the damn of the records contains programs. This time I play the cards so I bypass my management and talk with the CEO. It took 25 minutes of explaining what the problem was and he goes ballistic and he calls the VPs' in and reads them the riot act and by gosh that is what it took. This time we got the pats on the back and $$ as a bonus. I did not like how the bonus's were divvied up but did not say anything. Politics again trumps all.

Ed

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html

Reply via email to