Good benchmarks for testing GC performance/scalability

[Igor Veresov]

While SPECjvm2008 does not absolutely require a large heap - it never hurts to have plenty of memory. Also, in most subbenchmarks are structured as a bunch of parallel threads that operate on mostly thread-local data. Which makes it a perfect candidate for MPO. I've seen non-trivial gains with the NUMA allocator that is currently in Parallel Scavenger on very-NUMA machines, like multi-socket Mangy-Cours, back in the day.

igor

On Jun 11, 2012, at 11:01 PM, Lokesh Gidra wrote:

> Actually i have tried SPECjvm2008 but even there benchmarks do not have large heap requirements.
> 
> Is it worth to make both SPECjbb2012 and SPECjbb2005 tests or the former one covers the later?
> 
> Thanks
> Lokesh
> 
> Igor Veresov <iggy.veresov at gmail.com> wrote:
> If you have access to SPEC benchmarks,  SPECjbb2012 and SPECjvm2008 are worth trying too. They're all multithreaded and in many instances would benefit from memory placement opts.
> 
> igor
> 
> On Jun 11, 2012, at 7:39 PM, Lokesh Gidra wrote:
> 
>> Hello,
>> 
>> I am making some NUMA related optimizations to the parallelScavenge GC. And I want to test the performance on NUMA multo-core server with some multi-threaded benchmarks with multi-GB heap requirements. I have tried DaCapo but its heap requirements are pretty small. I know of SPECjbb2055, but nothing other than that. Can someone please suggest me some benchmarks for this purpose.
>> 
>> Thanks in advance.
>> 
>> Lokesh
>