PrintGCApplicationConcurrentTime / PrintGCApplicationStoppedTime
Y. Srinivas Ramakrishna
Y.S.Ramakrishna at Sun.COM
Wed Apr 15 09:27:53 PDT 2009
If you do extend it to all safepoints, please consider adding an
event time-stamp with the non-GC events which currently do not
carry a timestamp.
Otherwise, as David said, I hope all of that information can be
obtained from PrintSafepointstatistics.
Jon Masamitsu wrote:
> Question: Do users care about applications times output
> for PrintGCApplicationConcurrentTime only relative to
> I have a CR
> 6782663: Data produced by PrintGCApplicationConcurrentTime and
> PrintGCApplicationStoppedTime is not accurate
> where the complaint is that the application time as output
> for PrintGCApplicationConcurrentTime does not match the time
> as measured by the time between GC timestamps. Actually the
> user is adding up all the times reported for
> PrintGCApplicationConcurrentTime between the GC timestamps
> and saying that that sum can be vastly off from the
> time between GC timestamps. And the user is right.
> I think the problem is that the timers used to report
> PrintGCApplicationConcurrentTime are updated more often
> than the PrintGCApplicationConcurrentTime time is
> reported. In VMThread::loop() in share/vm/runtime/vmThread.cpp
> around line 425 the PrintGCApplicationConcurrentTime is
> reported before the call to SafepointSynchronize::begin().
> Whereas near line 391 and near line 520 calls to
> SafepointSynchronize::begin() do not report for
> PrintGCApplicationConcurrentTime. The calls to
> SafepointSynchronize::begin() will update the application
> timer (_app_timer via a call to
> RuntimeService::record_safepoint_begin()). Updating
> resets the timer to the current time and the
> PrintGCApplicationConcurrentTime output is then the
> time since the last safepoint (not since the application
> restarted after the last GC).
> So anyone know why the PrintGCApplicationConcurrentTime output
> does not print for all safepoints? Should it only printout
> when a VM operation is executed as it does now?
> If yes, should the spelling of
> PrintGCApplicationConcurrentTime be changed to drop
> the GC, PrintApplicationConcurrentTime. Similarly with
> PrintGCApplicationStoppedTime -> PrintApplicationStoppedTime.
> Or should it printout for all safepoints? This is
> simpler in that the printing could be added to
> RuntimeService::record_safepoint_begin()) so
> we would not miss new safepoints. But we might
> be dumping useless information.
> Or I could hack the code so that information is only
> printed around GC's. And maybe not printout some
> useful information.
More information about the hotspot-runtime-dev