Re: [PATCH] fix readahead pipeline break caused by block plug

[Andrew Morton]

On Wed, 1 Feb 2012 15:10:17 -0500
Vivek Goyal <vgoyal@xxxxxxxxxx> wrote:

> On Wed, Feb 01, 2012 at 04:18:07AM -0500, Christoph Hellwig wrote:
> > On Tue, Jan 31, 2012 at 10:36:53PM -0500, Vivek Goyal wrote:
> > > I still see that IO is being submitted one page at a time. The only
> > > real difference seems to be that queue unplug happening at random times
> > > and many a times we are submitting much smaller requests (40 sectors, 48
> > > sectors etc).
> > 
> > This is expected given that the block device node uses
> > block_read_full_page, and not mpage_readpage(s).
> 
> What is the difference between block_read_full_page() and
> mpage_readpage().

block_read_full_page() will attach buffer_heads to the page and will
perform IO via those buffer_heads.  mpage_readpage() feeds the page
directly to the BIO layer and leaves it without attached buffer_heads.

> IOW, why block device does not use mpage_readpage(s)
> interface?

We've tried it in the past and problems ensued.  A quick google search
for blkdev_readpages turns up stuff like
http://us.generation-nt.com/answer/patch-add-readpages-support-block-devices-help-201462802.html

> Applying following patch improved the speed from 110MB/s to more than
> 230MB/s.

Yeah.  It should be doable - it would be a matter of hunting down and
squishing the oddball corner cases.

--
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/