Re: buffers vs. pages vs. kernel speed

Dr. Werner Fink (werner@suse.de)
Tue, 17 Jun 1997 11:31:44 +0200


> If I understand correctly the patch below, when freeing pages succeeds,
> all caches are tried in a circular manner.
>
> It seems to me that such an algorithm can make problems for allocations
> with order > 0, since it decreases the probability to use contiguous pages
> for caching a given resource.

I have the same opinion ... but this state fix of Matthias make the system
work under high stress ... no `Couldn't get free page'.

>
> We had a buddy balancing, now we have something that looks like communism.
> I'm not sure that works better. :)

IMHO maybe we need a meta balancing between both strategies ... on low system
load the older one is the better one ... on high system stress the other one
seems to be more promising.

BTW: I've don a TYPO in my patch ... in linux/mm/page_alloc.c the
line 206 should look like:

if (priority != GFP_NFS && priority != GFP_BUFFER)

> Gerard.

Werner