On Mon, Jun 03, 2013 at 07:15:58PM +0200, Peter Zijlstra wrote: > On Mon, Jun 03, 2013 at 11:20:32AM -0400, Johannes Weiner wrote: > > On Mon, Jun 03, 2013 at 10:25:33AM +0200, Peter Zijlstra wrote: > > > On Thu, May 30, 2013 at 02:04:06PM -0400, Johannes Weiner wrote: > > > > Previously, page cache radix tree nodes were freed after reclaim > > > > emptied out their page pointers. But now reclaim stores shadow > > > > entries in their place, which are only reclaimed when the inodes > > > > themselves are reclaimed. This is problematic for bigger files that > > > > are still in use after they have a significant amount of their cache > > > > reclaimed, without any of those pages actually refaulting. The shadow > > > > entries will just sit there and waste memory. In the worst case, the > > > > shadow entries will accumulate until the machine runs out of memory. > > > > > > > > > > Can't we simply prune all refault entries that have a distance larger > > > than the memory size? Then we must assume that no refault entry means > > > its too old, which I think is a fair assumption. > > > > Two workloads bound to two nodes might not push pages through the LRUs > > at the same pace, so a distance might be bigger than memory due to the > > faster moving node, yet still be a hit in the slower moving one. We > > can't really know until we evaluate it on a per-zone basis. > > But wasn't patch 1 of this series about making sure each zone is scanned > proportionally to its size? Only within any given zonelist. It's just so that pages used together are aged fairly. But if the tasks are isolated from each other their pages may age at different paces without it being unfair since the tasks do not contend for the same memory. > But given that, sure maybe 1 memory size is a bit strict, but surely we > can put a limit on things at about 2 memory sizes? That's what this 10/10 patch does (prune everything older than 2 * global_dirtyable_memory()), so I think we're talking past each other. Maybe the wording of the changelog was confusing? The paragraph you quoted above explains the problem resulting from 9/10 but which this patch 10/10 fixes. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>