Re: [PATCH 11/12] vmscan: Write out dirty pages in batch

[tytso]

On Tue, Jun 15, 2010 at 12:43:42PM +0100, Mel Gorman wrote:
> 
> I'll do this just to see what it looks like. To be frank, I lack
> taste when it comes to how the block layer and filesystem should
> behave so am having troube deciding if sorting the pages prior to
> submission is a good thing or if it would just encourage bad or lax
> behaviour in the IO submission queueing.
> 

I suspect the right answer is we need to sort both at the block layer
and either (a) before you pass things to the filesystem layer, or if
you don't do that (b) the filesystem will be forced to do its own
queuing/sorting at the very least for delayed allocation pages, so the
allocator can do something sane.  And given that there are multiple
file systems that support delayed allocation, it would be nice if this
could be recognized by the writeback code, as opposed to having btrfs,
xfs, ext4, all having to implement something very similar at the fs
layer.

							- Ted
--
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/