I dont think it matters which machine you run on, you just run for a longer time, and get more samples that way. I would suggest you run it on the same machine as production so you get true samples. The hardware can change, for example the MVCL instruction used to be slow on one machine, and was fixed on the next generation.
You can use APA ( and Strobe) for two purposes 1. Where is the CPU used? A hot spot could be caused by a very expensive operation being done frequently. Or could be very efficient code being called millions of times. One lesson is to stop doing expensive calls, and to cache data from expensive calls. ( I remember a CICS transaction that did the same DB2 query 1000 times! ... each program checked the status of a flag in a table. They did the query once, and set a bit... CPU usage disappeared. APA showed this) 2. Where is the elapsed time spent. This could be disk I/O, Database waits/contention, CF etc Colin On Mon, 12 Apr 2021 at 21:37, Phil Smith III <li...@akphs.com> wrote: > This might be obvious, but if you can, you generally want to run APA on the > slowest system you own, because modern machines are too bloody fast: it can > have trouble identifying hotspots because it only goes to (I think) 100K > samples per second, and that's not enough. At least, that's what ISTR from > using it a few years ago-on then-current machines. Unless APA has increased > that granularity, this problem can only have gotten worse. > > > > I've heard of people considering a zPDT just so they could have a SLOWER > system to test on-when was the last time you heard that? > > > ---------------------------------------------------------------------- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN > ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
