[srosenberg]

So far my attempts to reproduce the alleged performance degradation have not been fruitful. I've written up a fairly detailed gist [1] on how to get CPU performance metrics; appendix also has a dump of C1 and C2 compiled methods (useful for comparison). I also ran on 2-node NUMA; binding cpu and memory to different nodes didn't yield a repro. either.

[1] https://gist.github.com/srosenberg/41611d5f40cfcbad51aa27eb0...

[cospaia]

I can't bind cpu and memory different.

  $ numactl --hardware
    available: 1 nodes (0)
  node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
  node 0 size: 15395 MB
  node 0 free: 2542 MB
  node distances:
  node   0 
    0:  10 


  $ numactl --cpunodebind=0 --membind=1 java PrimeSieve
  libnuma: Warning: node argument 1 is out of range

[cospaia]

I love the detail with which you describe the experiments. Let's see what I get running this on my Linux box!