I have T2000 8core 1Ghz.
I''ve got 145732 ops/s while running SPECjbb.
But only 702 ops/s if I will run simple dtrace script:
#!/usr/sbin/dtrace -s
dvm$target:::gc-start
{ self->ts=vtimestamp; }
dvm$target:::gc-finish
{ printf("GC ran for %d msec\n", (vtimestamp -
self->ts)/1000000 ); }
The impact is huge. Should it be so?
JVM is latest 6beta2. Solaris 10 is installed by default without any tuning.
This message posted from opensolaris.org
Hi Roman, dvm probes are implemented in so called Serviceability agents (dvmti, dvmpi). These agents uses JVMTI (or JVMPI) interfaces to request VM events and provide DTrace probes in the callback code for those events. This is why you see performance impact. I would suggest to use hotspot:::gc-begin and hotspot:::gc-end probes which are part of 6beta2? -katya Roman wrote:>I have T2000 8core 1Ghz. >I''ve got 145732 ops/s while running SPECjbb. >But only 702 ops/s if I will run simple dtrace script: > >#!/usr/sbin/dtrace -s >dvm$target:::gc-start >{ self->ts=vtimestamp; } > >dvm$target:::gc-finish >{ printf("GC ran for %d msec\n", (vtimestamp - self->ts)/1000000 ); } > >The impact is huge. Should it be so? > >JVM is latest 6beta2. Solaris 10 is installed by default without any tuning. > > >This message posted from opensolaris.org >_______________________________________________ >dtrace-discuss mailing list >dtrace-discuss at opensolaris.org >
I switch from dvm to hotspot and results are great! Performance impact is near to zero. # dtrace -s hotspot_gc.d -p `pgrep java` dtrace: script ''hotspot_gc.d'' matched 10 probes CPU ID FUNCTION:NAME 18 47760 __1cbDVM_ParallelGCFailedAllocationEdoit6M_v_:gc-end GC ran for 1 msec 18 47758 __1cVVM_ParallelGCSystemGCEdoit6M_v_:gc-end GC ran for 2361 msec 18 47758 __1cVVM_ParallelGCSystemGCEdoit6M_v_:gc-end GC ran for 734 msec 3 47760 __1cbDVM_ParallelGCFailedAllocationEdoit6M_v_:gc-end GC ran for 0 msec 3 47760 __1cbDVM_ParallelGCFailedAllocationEdoit6M_v_:gc-end GC ran for 0 msec 5 47760 __1cbDVM_ParallelGCFailedAllocationEdoit6M_v_:gc-end GC ran for 0 msec 31 47760 __1cbDVM_ParallelGCFailedAllocationEdoit6M_v_:gc-end GC ran for 0 msec This message posted from opensolaris.org