RFR(M):8252103:parallel heap inspection for ParallelScavengeHeap(Internet mail)

linzang(臧琳) linzang at tencent.com
Tue Sep 1 04:53:34 UTC 2020



Hi David,
     My mistake, it should be 2x faster with this patch, Thanks for point it out!
      Raw data of Jmap histo (Average of 5 times run)
      Before: 685ms
      After(4threads): 356ms

BRs,
Lin

> On Sep 1, 2020, at 12:45 PM, David Holmes <david.holmes at oracle.com> wrote:
> 
> Hi Lin,
> 
>> On 1/09/2020 12:28 pm, linzang(臧琳) wrote:
>> Dear All,
>>         May I ask your help to review this change of adding parallel
>> heap inspection for PSS Heap?
>>         bug:   https://bugs.openjdk.java.net/browse/JDK-8252103
>>         webrev:
>> http://cr.openjdk.java.net/~lzang/jmap-8214535/8252103/webrev.00/
>>         My preliminary measurement shows about ~1x speed up for when
>> using 4 threads of jmap -histo. (on a 4 core Linux machine, with -xmx8g
>> and objects takes 1.5GB in heap.)
> 
> ~1x speed up is not a speed up. What exactly did you mean here?
> 
> Thanks,
> David
> -----
> 
>>         As spaces in youngGen of parallelScvangeHeap is hard to be
>> divided for parallel iteration, this change just assign different worker
>> thread to process different space in youngGen, that is, 1 thread to
>> process whole eden space etc. so this may be one reason that speedup is
>> not scale with parallel thread number.
>> BRs,
>> Lin
> 


More information about the serviceability-dev mailing list