Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-8013895

G1: G1SummarizeRSetStats output on Linux needs improvement

    Details

    • Type: Bug
    • Status: Resolved
    • Priority: P4
    • Resolution: Fixed
    • Affects Version/s: hs24, hs25
    • Fix Version/s: hs25
    • Component/s: hotspot
    • Labels:
    • Subcomponent:
      gc
    • Resolved In Build:
      b36
    • CPU:
      x86
    • OS:
      linux

      Backports

        Description

        Here's how one set of stats looks like:
         Concurrent RS processed -1761455529 cards
          Of 9224262 completed buffers:
              9224262 (100.0%) by conc RS threads.
                    0 ( 0.0%) by mutator threads.
          Conc RS threads times(s)
                  0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
          Total heap region rem set sizes = 1120947K. Max = 4047K.
          Static structures = 600K, free_lists = 5312K.
            717820015 occupied cards represented.
            Max size region = 378:(O)[0x00000004fa000000,0x00000004fb000000,0x00000004fb000000], size = 4048K, occupied = 37303K.
            Did 209966 coarsenings.

        There are multiple issues as listed:
        1) -ve number for cards
        2) Conc RS threads times are always 0.00
        3) In-spite of specifying a period for stats (in this case -XX:G1SummarizeRSetStatsPeriod=1), the data seems to be cumulative.
        Also, it seems that cards processed by mutator threads are always 0 (0.0%).

        For #2, I understand that it is OS related, but, I'd appreciate any information, doesn't have to be exact (since per thread times are listed, so helps with performance analysis to see how many refinement threads were used during the specified period)
        For #3, I guess that the stats period was probably introduced so as not to pollute with excess data, but I take that to mean that the stats per period would be printed and not cummulative data.


          Attachments

            Issue Links

              Activity

                People

                • Assignee:
                  tschatzl Thomas Schatzl
                  Reporter:
                  mbeckwit Monica Beckwith
                • Votes:
                  0 Vote for this issue
                  Watchers:
                  6 Start watching this issue

                  Dates

                  • Created:
                    Updated:
                    Resolved: