Why you shouldn't be afraid to use VARCHAR in RPGLE...

I think I've maybe mentioned this in replies over the years. But as I
posted "Why you shouldn't be afraid to use VARCHAR in RPGLE" as a reply to
a post on the midrange list, I thought the subject would be worthwhile as
its own post here.

Back in 2016, a performance assessment by IBM's performance consulting team
reported back to my company that a certain (home grown) XML parsing routine
"was taking 23% of all cpu used by interactive jobs." The PEX report IBM
collected even showed that that CPU was mostly being used on only 13 lines
of code in that routine.

I happened to see the report, and took a look at the routine in question.
What I saw was a bunch of the following:

select;
when %trim($level(x)) = 'something';
....
when $trim($level(x)) = 'something-else');
....
endsl;


Obviously, doing the same work over and over again is a waste.

So I simply changed level and a couple other related variables from CHAR to
VARCHAR. Ironically, I didn't need to use %trim() at all, since level was
originally loaded via:

$level(x) = %subst(chars:1:stringlen);

The next year's performance report noted that CPU consumption by our XML
parsing routine had dropped to less than 1% of Interactive CPU usage. The
sample size was somewhat small, so IBM also looked at the routine
explicitly and noted that CPU usage per call to the routine dropped by a
factor of 15. Additionally, my testing showed that response time per call
dropped from an average of 0.39s to 0.13s!

Not bad for a day's work....
Most of which was regression testing!

Now at the time, we had 7 or 8 big P50-tier POWER8 servers...each with
approximately 10,000 users. So, as you can imagine, interactive CPU usage
was and still is hugely important to us.

Morals of the story:
- don't be afraid of VARCHAR
- lots of %trim() is often a red flag

HTH,
Charles

PS. Strongly consider VARCHAR for new DB fields. Just make sure you
ALLOCATE them appropriately.