Re: 2.5 page cache improvement idea

[Gerrit Huizenga <gerrit@us.ibm.com>]

That change is platform specific, isn't it?  I thought there was also
a recent IA-64 patch in progress for the same thing, but I might
be mistaken.  I'm thinking that it would be useful if the machine
independent code supported kernel text replication as well as
shared/read-only text replication for user level applications.

gerrit

> > node traffic is relatively expensive.  As a result, wasting a small
> > number of physical pages on duplicate read-only pages cuts down node
> > to node traffic in most cases.  Many NUMA systems have a cache for
> > remote memory (some cache only remote pages, some cache local and remote
> > pages in the same cache - icky but cheaper).  As that cache cycles,
> > it is cheaper to replace read-only text pages from the local node
> > rather than the remote.  So, for things like kernel text (e.g. one of
> > the SGI patches) and for glibc's text, as well as the text of other
> 
> The mips64 port onto SGI o2000 uses kernel text replication, that has
> been part of 2.3/2.4 for a long time now. Is there another patch you
> are talking about here?
> 
> Kanoj
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/