Mercurial > hg > graal-jvmci-8
diff src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp @ 3322:567c87d484a0
7041501: NUMA: Expand the old gen more aggressively
Summary: Expand the old gen in bigger increments
Reviewed-by: jmasa
author | iveresov |
---|---|
date | Wed, 04 May 2011 15:08:44 -0700 |
parents | eda9eb483d29 |
children | c9ca3f51cf41 |
line wrap: on
line diff
--- a/src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp Fri Apr 29 09:11:03 2011 +0200 +++ b/src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp Wed May 04 15:08:44 2011 -0700 @@ -224,6 +224,12 @@ const size_t alignment = virtual_space()->alignment(); size_t aligned_bytes = align_size_up(bytes, alignment); size_t aligned_expand_bytes = align_size_up(MinHeapDeltaBytes, alignment); + + if (UseNUMA) { + // With NUMA we use round-robin page allocation for the old gen. Expand by at least + // providing a page per lgroup. Alignment is larger or equal to the page size. + aligned_expand_bytes = MAX2(aligned_expand_bytes, alignment * os::numa_get_groups_num()); + } if (aligned_bytes == 0){ // The alignment caused the number of bytes to wrap. An expand_by(0) will // return true with the implication that and expansion was done when it