cfq-iosched: don't let idling interfere with plugging
authorJens Axboe <jens.axboe@oracle.com>
Tue, 7 Apr 2009 09:38:31 +0000 (11:38 +0200)
committerJens Axboe <jens.axboe@oracle.com>
Tue, 7 Apr 2009 09:38:31 +0000 (11:38 +0200)
commitb029195dda0129b427c6e579a3bb3ae752da3a93
tree7f8752d2fefac40b0b25720cc1b0fbbd8aed173a
parent2385327725419a76cfbca7258abd95908b8ba9eb
cfq-iosched: don't let idling interfere with plugging

When CFQ is waiting for a new request from a process, currently it'll
immediately restart queuing when it sees such a request. This doesn't
work very well with streamed IO, since we then end up splitting IO
that would otherwise have been merged nicely. For a simple dd test,
this causes 10x as many requests to be issued as we should have.
Normally this goes unnoticed due to the low overhead of requests
at the device side, but some hardware is very sensitive to request
sizes and there it can cause big slow downs.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
block/cfq-iosched.c