Safepointing & Locking issue
mailing at nmichael.de
Thu Jul 8 11:04:52 PDT 2010
Thanks, I will add -XX:+TraceSafepointCleanupTime as well. The output
for -XX:+PrintSafepointStatistics was in my previous mail -- it's
basically "DeoptimizeFrame" about 1200 times a second.
And Ramki, you are right: It's actually more about the frequency of the
safepoints, not necessarily their duration. I didn't do a real (i.e.
SunStudio) profiling, just some quick DTrace profiling, which always
captured deflate_idle_monitors() being top of the stack.
Am 08.07.2010 19:03, schrieb Y. S. Ramakrishna:
> On 07/08/10 08:06, Karen Kinnear wrote:
>> It would also be very helpful if you could possibly gather information
>> to determine
>> why you are running so many safepoints.
>> Could you possibly try running with the following flags on a fairly
>> current jdk7 binary?
>> -XX:+PrintSafepointStatistics -XX:+TraceSafepointCleanupTime
>> That should tell you why you are running so many safepoints (which
>> would not be caused
>> by lock inflation/deflation). It will also tell you details of time
>> spent in different components
>> of the safepoint cleanup time, including lock deflation. Note that the
>> flags require a development
>> build, like fastdebug.
> The two you mention above are also available in product builds, thanks
> to Xiaobin.
> But probably moot now considering what Nicolas has figured out since the
> first email on this thread (i.e. frequency of safepoints, not their
> duration or deflation per se, being the cause of his overhead).
> -- ramki
More information about the hotspot-runtime-dev