We is using Webshpere on OraPeeps E811, 8.94, I5 with several LPAR (PRD on 1 LPAR) I don’t know if any of this applies to you or not. First, we changed the I5 so no resource sharing between LPAR, changed the Max Heap on Web App Server down from *NOMAX (despite the warnings), and increased the initial heap a little. Increased net processes to 16 from 4 in jde.ini. Our biggest gain came from the nomax change which seems to help the GC perform more efficiently or something even on total system overload.
Cleaning up the cache via SAW may help but it may just be coincidence (like for us)
We gots help from IBM on Websphere (OraPeeps does not do performance tuning by the way, surprise!)
Can you reproduce performance problems in testing?
Is it just interactive or do batch jobs run slowly?
Is it always slow or intermittent?
Are you monitoring your heap? (SAW System Summary)
Are you monitoring your GCs?
Do you have any vacation coming up?
We reproduced the problem in test, played with some settings, reproduced probs in production after hours and then applied WAS changes then could not reproduce probs in production (like locking the entire system up for 15 minutes while the nomax heap cleared up???) Since our problems have been intermittent we are still skeptical, but since we started tuning things we have had much better response.
Also, we have noticed far too often that the package build process has been running during some of our user complaints of either performance or error messages and since our tuning this has become more apparent.
Good Luck