Uploaded image for project: 'JikesRVM'
  1. RVM-391

Assign CPU # to each VM_Processor to have more predictable performance results.

    Details

    • Type: Improvement
    • Status: Closed
    • Priority: Lowest
    • Resolution: Fixed
    • Affects Version/s: None
    • Fix Version/s: 2.9.3
    • Labels:
      None
    • Environment:

      The patch works on ia32-linux with NPTL

      Description

      Some recent hardware has asymmetric cache layout.
      For example, Intel Core 2 Quad has 4 cores, but only 2 cores in pair share L2.
      Thus the performance in multithreaded benchmarks shows high variance depending on
      which core the VM_Processor is mapped to.
      The patch assigns CPU number to the pthreads to have more predictable results.
      Any users can only modify the userlevel scheduler to do further optimizations for cache affinity studies.

        Attachments

          Activity

            People

            • Assignee:
              ianrogers Ian Rogers
              Reporter:
              habals Jungwoo Ha
            • Votes:
              0 Vote for this issue
              Watchers:
              0 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved: