RFR (S): 8204081: Mismatch in rebuild policy and collection set chooser causes remembered sets to be kept errorneously

Stefan Johansson stefan.johansson at oracle.com
Thu May 31 08:35:57 UTC 2018



On 2018-05-30 14:58, Thomas Schatzl wrote:
> Hi all,
> 
>    can I have reviews for this small change that causes some performance
> loss after earlier rebuild remembered set changes by keeping around
> remembered sets for some regions for too long?
> 
> The problem is that the remembered set tracker and the collection set
> chooser have a different idea of what remembered sets should be kept.
> I.e. the remembered set tracker basically takes all regions that have a
> low enough occupancy, while the collection set chooser only looks at
> regions that have been marked in the last marking cycle, and so does
> not dispose of remembered sets for regions that have no marking
> information, but got added because they are low-occupancy due to other
> reasons (fragmentation).
> It also causes the collection set chooser to not consider these regions
> for evacuation until the next marking cycle.
> 
> This can save a few GCs in some applications, and seems to improve
> performance.
> 
> This change fixes this issue.
> 
> CR:
> https://bugs.openjdk.java.net/browse/JDK-8204081
> Webrev:
> http://cr.openjdk.java.net/~tschatzl/8204081/webrev/
Nice fix, looks good!
Cheers,
Stefan

> Testing:
> hs-tier1-3,jdk-tier1-2
> 
> Thanks,
>    Thomas
> 


More information about the hotspot-gc-dev mailing list