Request for review (s) - 8024366: UseNUMAInterleaving should be available for other GC's.

[Thomas Schatzl]

Hi,

On Mon, 2014-01-27 at 10:46 -0800, Jon Masamitsu wrote:
> On 01/27/2014 12:55 AM, Thomas Schatzl wrote:
> > Hi Jon,
> >
> >    the comment
> >
> > +     // For those collectors or operating systems that do
> >        // not support full UseNUMA, we will map to UseNUMAInterleaving for now
> >
> > indicates to me that the following mapping is done only for "those
> > collectors or operating systems that do not support full UseNUMA" while
> > it seems to me that this mapping is now done for all collectors/OSes
> > when UseNUMA is enabled. (Regardless of support, could not find any
> > logic that disables wrongly set UseNUMA).
> 
> Are you asking about OS's that have no support for anything
> related to NUMA?
> 
> What's an example where UseNUMA needs to be disabled?

UseNUMA does not make sense to enable e.g. on a single socket machine,
or more generally when the VM process is bound to a set of hardware
threads that are all located on a single NUMA node. Or an OS without
NUMA support (OS X?).

I think it is fine to just rely on the user doing the right thing in all
cases (disabling UseNUMA manually or not setting it in the first place).

The patch itself looks fine to me, my only concern were the comments
which seem to not reflect the code after the change.

Thanks,
Thomas