Huge pages

[Volker Simonis]

On Thu, Apr 11, 2013 at 3:38 AM, Tiago Stürmer Daitx <
tdaitx at linux.vnet.ibm.com> wrote:

> On Wed, 2013-04-10 at 19:26 +0200, Volker Simonis wrote:
> >
> > I've attached a small patch for you, which should enable UseHugeTLBFS
> > on our Linux/PPC port (i.e. you should be able to use -XX:
> > +UseHugeTLBFS -XX:+UseLargePages) with and without compressed
> > pointers.
>
> The patch works and seems to give a bit better performance than
> hugepages through shm... but still only for maximum heaps of 1984
> megabytes. I saw around 5% better run times with UseHugeTLBFS, while
> UseSHM was around 4% better. This is kind bellow the jitter I see
> sometimes, so don't take it for granted until I repeat the tests a
> significant amount of times.
>

Unfortunately I found out the patch is far from stable (it breaks
'-XX:-UseHugeTLBFS -XX:+UseLargePages -XX:+UseSHM'). So please use it with
caution.
I think it will be not trivial to get this working with all the
combinations of different GCs and +/- compressed oops on all Linux
architectures.

It would be really great if you could find sombody from the LTC who can
explain why mmap with MAP_HUGETLB behaves differently on x86 and ppc.


> The problem is that any heap bigger than 1984m gives ~7% worse
> performance compared to _not_ using hugepages - there was no performance
> loss when using shm.
>

This may come from being forced into a slower compressed oops mode.


> Please let me know how else I can help. =)
>
> Regards,
> Tiago Daitx
>
> --
> Tiago Stürmer Daitx
> tdaitx at linux.vnet.ibm.com
> IBM - Linux Technology Center
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/ppc-aix-port-dev/attachments/20130411/a255a580/attachment-0001.html