Messages in this thread | | | Date | Tue, 30 Jun 2009 22:13:10 -0400 | From | Rik van Riel <> | Subject | Re: Found the commit that causes the OOMs |
| |
KOSAKI Motohiro wrote:
> if my guess is correct, we need to implement #-of-reclaim-process throttling > mechanism.
There are probably some other things that want throttling, too.
For example, the number of pages currently under IO can be as large as the entire file and anon inactive lists, which can cause page reclaim to fail because none of the pages are reclaimable yet.
This is probably not a big issue for the page cache, since the readahead window will collapse before we hit this problem.
However, we may want to take measures to ensure that the total number of pages in swap readahead do not take up the entire inactive anon list - maybe we should limit it to half that amount, to stay on the safe side?
I'll whip up a patch for this tomorrow.
That should get rid of the OOMs that have been observed with the swap readahead patches by Johannes.
-- All rights reversed.
| |