[RFCv2][PATCH 0/5] mm: Batch page reclamation under shink_page_list

From: Dave Hansen
Date: Thu May 16 2013 - 16:34:31 EST

These are an update of Tim Chen's earlier work:


I broke the patches up a bit more, and tried to incorporate some
changes based on some feedback from Mel and Andrew.

Changes for v2:
* use page_mapping() accessor instead of direct access
to page->mapping (could cause crashes when running in
to swap cache pages.
* group the batch function's introduction patch with
its first use
* rename a few functions as suggested by Mel
* Ran some single-threaded tests to look for regressions
caused by the batching. If there is overhead, it is only
in the worst-case scenarios, and then only in hundreths of
a percent of CPU time.

If you're curious how effective the batching is, I have a quick
and dirty patch to keep some stats:



To do page reclamation in shrink_page_list function, there are
two locks taken on a page by page basis. One is the tree lock
protecting the radix tree of the page mapping and the other is
the mapping->i_mmap_mutex protecting the mapped pages. This set
deals only with mapping->tree_lock.

Tim managed to get 14% throughput improvement when with a workload
putting heavy pressure of page cache by reading many large mmaped
files simultaneously on a 8 socket Westmere server.

I've been testing these by running large parallel kernel compiles
on systems that are under memory pressure. During development,
I caught quite a few races on smaller setups, and it's being
quite stable that large (160 logical CPU / 1TB) system.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/