On Mon, 2007-03-19 at 20:52 -0400, Rik van Riel wrote:Split the anonymous and file backed pages out onto their own pageout
queues. This we do not unnecessarily churn through lots of anonymous
pages when we do not want to swap them out anyway.
This should (with additional tuning) be a great step forward in
scalability, allowing Linux to run well on very large systems where
scanning through the anonymous memory (on our way to the page cache
memory we do want to evict) is slowing systems down significantly.
This patch has been stress tested and seems to work, but has not
been fine tuned or benchmarked yet. For now the swappiness parameter
can be used to tweak swap aggressiveness up and down as desired, but
in the long run we may want to simply measure IO cost of page cache
and anonymous memory and auto-adjust.
We apply pressure to each of sets of the pageout queues based on:
- the size of each queue
- the fraction of recently referenced pages in each queue,
not counting used-once file pages
- swappiness (file IO is more efficient than swap IO)
Please take this patch for a spin and let me know what goes well
and what goes wrong.
Rick: Which tree is the patch against. Diffs say 2.6.20.x86_64, but
doesn't apply to 2.6.20 which doesn't use __inc_zone_state() for things
like nr_active, nr_inactive, ...
Also, in the snippet:
--- linux-2.6.20.x86_64/mm/swap_state.c.vmsplit 2007-02-04-0400
13:44:54.000000000 -0500
+++ linux-2.6.20.x86_64/mm/swap_state.c 2007-03-19 12:00:23.000000000
@@ -354,7 +354,7 @@ struct page *read_swap_cache_async(swp_e
/*
* Initiate read into locked page and return.
*/
- lru_cache_add_active(new_page);
+ lru_cache_add_anon(new_page);
swap_readpage(NULL, new_page);
return new_page;
}
Should that be lru_cache_add_active_anon()? Or did you intend to add it
to the inactive anon list?
Finally, could you [should you?] skip scanning the anon lists--or at
least the inactive anon list--when nr_swap_pages == 0? The anon pages
aren't going anywhere, right? I think this would obviate Christoph L's
patch to exclude anon pages from the LRU when there is no swap.