[Deploying Sakai] Strange Java default NewSize behavior
da1 at vt.edu
Thu Sep 24 10:24:09 PDT 2009
While troubleshooting some ultimately unrelated problems earlier this week, we ran jstat to check how our VM memory was doing, and we found that the new generation space (whose size we did not then specify) was frighteningly tiny. Out of our 6GB heap, Java was using only about 40MB for the entire new generation, which meant multiple minor GCs every second, and a lot of junk in Old.
This was happening on all three of our production servers. On our test server, using identical settings (same Java, same OS, same Sakai build, same Tomcat, etc etc etc), Java was choosing a sane value for new (2GB or so). The machines are both Dell 1950s with dual quad-core 64-bit Xeon processors and 8GB of RAM. The test machine is 10 months older, so I'm sure there was a processor revision in between, but otherwise they are identical.
We ended up just specifying NewSize in JAVA_OPTS to address the problem, but has anyone run into anything like this before?
Director of Systems Integration and Support
Virginia Tech Learning Technologies
More information about the production