[Beowulf] Re: Feedback on large pages in Linux (hahn@physics.mcmaster.ca)

Greg Lindahl greg.lindahl at qlogic.com
Fri Jul 28 22:29:38 PDT 2006


On Wed, Jul 26, 2006 at 06:55:29PM -0400, Robert G. Brown wrote:

> Isn't there a fairly general rule of thumb, that systems that get a
> large benefit from streaming I/O through a cache of any size pay a
> penalty in terms of repeated flushes for sufficiently NON-streaming
> (random or variable stride) I/O?

Uh, yeah. However, in a system with multiple page sizes, nobody says
that you have to use the large size for your file mmaps or I/O.

Paging is also less efficient with large pages, but modern systems
don't page much.

BTW, you should warn me when you invoke my name at the end of a long
post, it's not like I'm as omniscient as Kibo...

> (working for a compiler company, after all;-)...

... did a lot more work on the interconnect. Maybe I just suffer from
Male Answer Syndrome?

-- greg



More information about the Beowulf mailing list