Re: [PATCH] mm: exempt pcp alloc from watermarks

[Nick Piggin <nickpiggin@yahoo.com.au>]

Christoph Lameter wrote:
> Could we simply try to get a page from the pcp of the best zone before 
> doing any other processing? That way we may actually improve the performance 
> of alloc pages.

There is very little other processing in my patch. The only thing is
really the cpuset check, which really seems (IMO) to be required.

I don't expect performance to change much right now... scalability will be
improved for CPUs allocating from the same zones (due to batched access of
zone data), but that shouldn't be too common on machines these days.

I expect the bigger win will come from not having to cycle through a huge
number of zones *every* allocation when memory is low, in Andrew's multi
node container situation[*]. This won't work if you only check the best
zone.

[*] If it is still a problem, they should be able to increase the pcp batch
size until it isn't.

> Could we inline the attempt to get a page from the pcp?

Not sure. My instinct is not to inline unless it makes the code smaller,
provides obvious constant folding benefits, or runs faster. The former
two don't seem to be the case, and the latter is probably not measurable.

-- 
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com 

--
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-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>