Re: Profiling RPGLE programs

I'm sure a lot of it will end up being I/O related as there are many levels of ancillary files/records associated with each detail being processed. Hoping it could be as simple as changing the LF so similar records get grouped to leverage caching.

Roger Harman
COMMON Certified Application Developer - ILE RPG on IBM i on Power

--

From: MIDRANGE-L <midrange-l-bounces@xxxxxxxxxxxxxxxxxx> on behalf of Rob Berendt <rob@xxxxxxxxx>
Sent: Wednesday, February 19, 2020 9:37 AM
To: Midrange Systems Technical Discussion <midrange-l@xxxxxxxxxxxxxxxxxx>
Subject: RE: Profiling RPGLE programs

Basically the one time I rewrote a program based solely on performance was by measuring the I/O's.  Then again, not a whole lot of people use RPG for nuclear theory or weather projection.

In my case the I/O's were very high.  Huge detail file needed to be summarized to a fresh summary file.  Original program read every row, checked summary file for a grouping row, if there wasn't one write it.  If there was a grouping row update it.

Rewrite only wrote out to summary file at control break time.

There were many detail rows to each summary row.

We also journal all files so this cut way back on the number of journal entries too.

I remember it chopped it down from a few hours to minutes.

Rob Berendt