Running Vista 4.1.2 this summer, we saw certain activity grabbed memory that was never released back into the pool. It only made our servers unresponsive twice, and the activity associated with the absorption of the entire 1.5 gig allocated memory turned out to be garbage collection. So we turned it off. Other institutions with bigger installations haven’t been quite so fortunate.
But it’s made me paranoid. Now I’m on the BEA Weblogic Console a few minutes every day watching JVM usage. In the new 9.2 console, I select Environment, Servers, NodeA in the left hand panel. Then across the horizontal tabs in the right frame, Monitoring, JMS, Performance. Although I don’t have the nice graph BEA Weblogic 8.1 had, I can click a refresh button which sits there and refreshes the stats every 10 seconds or so (I have no idea where its frequency is set, do you?).
If I’m on the back end at the same time, I can kind of get a feel for what activity is going on because I have seen Node A’s free memory decrease right down to %11 of allocated. After 5 minutes of it sitting there, my paranoia gets the best of me and I force a garbage collection (also on that page).
When start of the semester activity dies down a little further, my goal is to make better correlations between the memory usage and what the ‘big tasks’ going on at the moment are. I can report that garbage collection is no longer the big one. It runs at high noon every day between 2 and 7 minutes and finishes with the "Succeeded" message.