Continous FULL GC

Question

We have an application running on IBM websphere server with the following JVM and heap settings
Heap Settings
Min Hep : 750MB, MaxHeap : 1500 MB
JVM Settings
-XX:PermSize=64m -XX:MaxPermSize=128m -XX:+PrintGCTimeStamps -XX:+PrintHeapAtGC
In the verbose log file we observed that FULL GC is triggered even though we have enough memory in all generations (young,old, permannent) of JVM. Here is a sample output from the GC log file
} {Heap before GC invocations=1420:
Heap
def new generation   total 83136K, used 217K [0x94400000, 0x997a0000, 0x9eaa0000)
  eden space  80640K,   0% used [0x94400000, 0x94436570, 0x992c0000)
  from space   2496K,   0% used [0x99530000, 0x99530000, 0x997a0000)
  to   space   2496K,   0% used [0x992c0000, 0x992c0000, 0x99530000)
tenured generation   total 682688K, used 115942K [0x9eaa0000, 0xc8550000, 0xf2000000)
   the space 682688K,  16% used [0x9eaa0000, 0xa5bd9a08, 0xa5bd9c00, 0xc8550000)
compacting perm gen  total 131072K, used 40314K [0xf2000000, 0xfa000000, 0xfa000000)
   the space 131072K,  30% used [0xf2000000, 0xf475ebc0, 0xf475ec00, 0xfa000000)
64557.286:[b] [Full GC 116159K->115961K(765824K), 2.1373354 secs] Heap after GC invocations=1421:
Heap
def new generation total 83136K, used 0K [0x94400000, 0x997a0000, 0x9eaa0000)
eden space 80640K, 0% used [0x94400000, 0x94400000, 0x992c0000)
from space 2496K, 0% used [0x99530000, 0x99530000, 0x997a0000)
to space 2496K, 0% used [0x992c0000, 0x992c0000, 0x99530000)
tenured generation total 682688K, used 115961K [0x9eaa0000, 0xc8550000, 0xf2000000)
the space 682688K, 16% used [0x9eaa0000, 0xa5bde658, 0xa5bde800, 0xc8550000)
compacting perm gen total 131072K, used 40314K [0xf2000000, 0xfa000000, 0xfa000000)
the space 131072K, 30% used [0xf2000000, 0xf475ebc0, 0xf475ec00, 0xfa000000)
We started observing this behaviour only after we started using webservice (HTTP) as communication protocal instead of RMI/IIOP. Is there any specific reason for this kind of bahaviour. Do we need to make any special settings to avoid this behavior.

Answer 1

This is more related to websphere than to JVM. It seems websphere directly call Full GC, because JVM itself won't do Full GC unless low on memory.
You can try to change GC mode to reduce pause/improve performance.