RFR: 8027675: Full collections with Serial slower in JDK 8 compared to 7u40

[Thomas Schatzl]

Hi,

On Fri, 2013-11-22 at 09:56 +0100, Mikael Gerdin wrote:
> Coleen,
> 
> On November 21, 2013 8:11:22 PM Coleen Phillimore 
> <coleen.phillimore at oracle.com> wrote:
> >
> > Stefan,
> >
> > I think the function MarkSweep::adjust_class_loader() is now unused, but I 
> > couldn't find where we adjust the class in the CLD walk. There are similar 
> > functions in ParallelScavenge - are they also not needed?   I don't 
> > recommend adding this change, just noting it may be cleaned up later if so.
> 
> All Klasses are adjusted because we call oops_do on CLDG with the adjust 
> closure. We don't need to discover the CLDs a second time after we are done 
> with marking.
> 
> MarkSweep::adjust_class_loader() can be removed since it's not used.
> The same problem affects ParallelScavenge but it's possible that the 
> regression is not as obvious due to the parallelism.

Stefan, could you file an RFE for this?

> >
> > This change seems good but I'm not a GC person.
> 
> Stefan,
> This change looks good but since I was involved with developing the fix I'd 
> like another gc Reviewer to look at the fix as well.

Looks good to me (also the .01 webrev).

Thomas