[Beowulf] Feedback on large pages in Linux

hahn at physics.mcmaster.ca hahn at physics.mcmaster.ca
Tue Jul 25 15:15:42 PDT 2006

> Can someone explain me why 2MB can be faster than 16MB?

suppose you touch data with stride 2MB.  if all you use is 2M pages,
you start taking misses after only 8 references.

I think the take-home message is that bigpages help if you have 
some large, contiguous, hot regions.  but it seems to me that blocking
would give rise to such, especially for code which attempts to 
tune for multi-MB caches.

More information about the Beowulf mailing list