java - PermGen thrashing at 99%, but nowhere near MaxPermSize -

i have repeatable situation jvm undergoing heavy gc load. when request jvm stats using jmap -heap, info below (this oracle jdk 1.7.0_25 on linux)

note while says maxpermsize 256m, says permgen sitting @ 136mb @ 99.9% capacity.

that explain gc thrashing, question why isn't jvm expanding permgen full available 256m? there parameter can prevent pool expansion happening, , stopping jvm making full use of 256m?

note tenured pool getting bit tight, near as permgen.

using thread-local object allocation. mark sweep compact gc  heap configuration:    minheapfreeratio = 5    maxheapfreeratio = 10    maxheapsize      = 805306368 (768.0mb)    newsize          = 1048576 (1.0mb)    maxnewsize       = 4294901760 (4095.9375mb)    oldsize          = 4194304 (4.0mb)    newratio         = 8    survivorratio    = 8    permsize         = 50331648 (48.0mb)    maxpermsize      = 268435456 (256.0mb)    g1heapregionsize = 0 (0.0mb)  heap usage: new generation (eden + 1 survivor space):    capacity = 43909120 (41.875mb)    used     = 495240 (0.47229766845703125mb)    free     = 43413880 (41.40270233154297mb)    1.1278750291511195% used eden space:    capacity = 39059456 (37.25mb)    used     = 495240 (0.47229766845703125mb)    free     = 38564216 (36.77770233154297mb)    1.2679132039114933% used space:    capacity = 4849664 (4.625mb)    used     = 0 (0.0mb)    free     = 4849664 (4.625mb)    0.0% used space:    capacity = 4849664 (4.625mb)    used     = 0 (0.0mb)    free     = 4849664 (4.625mb)    0.0% used tenured generation:    capacity = 389492736 (371.44921875mb)    used     = 350542912 (334.30377197265625mb)    free     = 38949824 (37.14544677734375mb)    89.99985868799361% used perm generation:    capacity = 143392768 (136.75mb)    used     = 143338624 (136.6983642578125mb)    free     = 54144 (0.0516357421875mb)    99.96224077353747% used  174149 interned strings occupying 19526656 bytes. 

one possible reason computer has no more available memory, hence jvm can't allocate more space permgen (you need ~ +30% of available memory 1 configured).

other reason (and i'm more one) weird behaviour of java ergonomics. may possible jvm decide permgen "best size" 136mb failed in increasing @ point. maybe because of memory leaks caused permgen. can try use -xx:+cmsclassunloadingenabled -xx:+cmspermgensweepingenabled if use concurrent mark , sweep gc.