There is a huge difference in memory consumption when you run SOA11g (PS4 in my case). The screens are based on a full-production clustered environment. One node is running with sub-optimal settings, while the other node is running optimal settings. A load-balancer is equally spreading the requests to the nodes.
Running SOA 11g on VMWare, 1 CPU and 3GB memory, with the options:
-Xms1280m
-Xmx1280m
-XX:MaxPermSize=512m
-XX:+UseParallelGC
Running SOA 11g on VMWare, 1 CPU and 3GB memory, with the options:
-Xms1280m
-Xmx1280m
-XX:MaxPermSize=512m
-Xss256k
-XX:+UseConcMarkSweepGC
-XX:+UseParNewGC
-XX:+CMSIncrementalMode
-XX:+CMSIncrementalPacing
The last XX options shows that the memory consumption is dramaticlly lower and garbage collection is lower but frequenter.