[Beowulf] Feedback on large pages in Linux

[Greg Lindahl]

On Fri, Jul 21, 2006 at 11:39:23PM -0400, hahn at physics.mcmaster.ca wrote:

> it would be incredibly useful if the HPC community could come up with 
> some actual numbers on how important bigpages are.  I'm ashamed to say
> that I haven't made any effort to measure this on any of my clusters.
> oprofile appears to make it pretty easy to capture real data on how 
> often TLB's are a problem.

Actually it's not hard to measure the benefit -- take an OS like
Solaris and use Sun's SpecOMP runs with and without huge pages...
also spec_rate...  insomuch as you ever see their SunStudio compilers
beating ours, huge pages are involved. But I've never seen a study
break out the benefits separately, they're always Linux & small pages
& compiler A vs. not Linux & large pages & compiler B...

-- greg