On Mon, Feb 12 2007, Rusty Russell wrote: > On Mon, 2007-02-12 at 06:32 +0100, Jens Axboe wrote: > > On Mon, Feb 12 2007, Rusty Russell wrote: > > > On Mon, 2007-02-12 at 05:43 +0100, Jens Axboe wrote: > > > > Here you map the entire request (lets call that segment A..Z), but > > > > end_request() only completes the first chunk of the request. So > > > > elv_next_request() will retrieve the same request again, and you'll then > > > > map B..Z and repeat that transfer. So unless I'm missing some other part > > > > here (just read it over quickly), you are re-doing large parts of a > > > > merged request several times. > > virtbench before: > Time to read from disk (256 kB): 18654562 nsec > After: > Time to read from disk (256 kB): 8018468 nsec > > Thanks Jens!! My pleasure, it's not often you get to make that big a performance improvement with just a little few lines of change :-) I guess you'll take changes to make this driver queuing as well? It's pretty important for good guest io performance as well. -- Jens Axboe
