On Fri, Oct 05, 2007 at 08:30:28PM +0800, Fengguang Wu wrote:
quoted text
> The improvement could be:
> - kswapd is now explicitly preferred to do the writeout;

Careful. kswapd is much less efficient at writeout than pdflush
because it does not do low->high offset writeback per address space.
It just flushes the pages in LRU order and that turns writeback into
a non-sequential mess. I/O sizes decrease substantially and
throughput falls through the floor.

So if you want kswapd to take over all the writeback, it needs to do
writeback in the same manner as the background flushes. i.e.  by
grabbing page->mapping and flushing that in sequential order rather
than just the page on the end of the LRU....

I documented the effect of kswapd taking over writeback in this
paper (section 5.3):

http://oss.sgi.com/projects/xfs/papers/ols2006/ols-2006-paper.pdf

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group
-
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/