On Wed, Apr 15, 2009 at 01:00:14PM -0700, Greg Lindahl wrote:
> On Tue, Apr 14, 2009 at 03:44:17PM -0700, Christian Bell wrote:
>
> > Sure, capacity and conflict misses will apply like
> > they do in any cache but Jon's statement remains correct -- in SPMD
> > applications, you really want all processes to benefit from cache hits in
> > your "single program".
>
> ... then why do vendors make such a fuss about duplicating program
> text pages in big NUMA machines?
>
> As an example:
>
> www.sc2001.org/papers/pap.pap241.pdf
Good link -- still the key is in the phrase "big NUMA machines"
of the previous post.
i.e. NUMA "non uniform memory access".
Replication of interesting pages can flatten the access times
across the machine and partly remove the Non from NUMA ---> UMA.
> The issue is that fetching from distant memory is expensive, even if
> there's no writing going on. Locality, locality, locality.
Exactly.
> -- greg
Later,
mitch (who should have read this thread a couple days ago, all the canapes are
cold)
--
T o m M i t c h e l l
Found me a new hat, now what?
_______________________________________________
Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
