RFR(XS): 8015244: G1: Verification after a full GC is incorrectly placed

Bengt Rutisson bengt.rutisson at oracle.com
Fri May 24 13:02:15 UTC 2013


Hi John,

The change looks good.

Thanks for running the simple performance test! That was a good idea! 
And I think a 4% the overhead seems reasonable.

Bengt

On 5/23/13 1:20 AM, John Cuthbertson wrote:
> Hi Everyone,
>
> Can I have a couple of volunteers look over this small fix? The webrev 
> can be found at: http://cr.openjdk.java.net/~johnc/8015244/webrev.0/
>
> Summary:
> While investigating some additional verification code for "7145569: 
> G1: optimize nmethods scanning", I realized that the verification that 
> occurs after a full GC is misplaced. I think it should come after the 
> rebuilding of the RSets and verify the rebuilt RSets. Verifying the 
> RSets before the full GC is less interesting and potentially wasteful 
> so I placed the RSet verification during full GCs under control of an 
> additional diagnostic flag.
>
> These aren't true performance numbers but when I ran GCOld with the 
> new flag enabled (1G heap), I saw an increase in full GC times of 
> around 4% on my workstation (VerifyBefore and VerifyAfter were both 
> enabled).
>
> Testing:
> GCOld; CG test suite
>
> Thanks,
>
> JohnC




More information about the hotspot-gc-dev mailing list