On Fri, 29 Jun 2012, Andrew Morton wrote:
Ok, the standard deviation increases when merging pages from nodes withI've tested this patch on numa machines with 2, 4 and 8 nodes andooh, numbers! Thanks.
measured speed of memory access inside of KVM guests with memory pinned
to one of nodes with this benchmark:
http://pholasek.fedorapeople.org/alloc_pg.c
Population standard deviations of access times in percentage of average
were following:
merge_nodes=1
2 nodes 1.4%
4 nodes 1.6%
8 nodes 1.7%
merge_nodes=0
2 nodes 1%
4 nodes 0.32%
8 nodes 0.018%
remote distance, that makes sense. But if that's true, then you would
restrict either the entire application to local memory with mempolicies or
cpusets, or you would use mbind() to restrict this memory to that set of
nodes already so that accesses, even with ksm merging, would have
affinity.