Re: cfq-iosched: tiobench regression

[Corrado Zoccolo <czoccolo@gmail.com>]

Hi Shaohua,
On Thu, Dec 24, 2009 at 1:55 AM, Shaohua Li <shaohua.li@intel.com> wrote:
> df5fe3e8e13883f58dc97489076bbcc150789a21
> b3b6d0408c953524f979468562e7e210d8634150
> The coop merge is too aggressive. For example, if two tasks are reading two
> files where the two files have some adjecent blocks, cfq will immediately
> merge them. cfq_rq_close() also has trouble, sometimes the seek_mean is very
> big. I did a test to make cfq_rq_close() always checks the distence according
> to CIC_SEEK_THR, but still saw a lot of wrong merge. (BTW, why we take a long
> distence far away request as close. Taking them close doesn't improve any thoughtput
> to me. Maybe we should always use CIC_SEEK_THR as close criteria).
Yes, when deciding if two queues are going to be merged, we should use
the constant CIC_SEEK_THR.
> So sounds we need make split more aggressive. But the split is too lazay,
> which requires to wait 1s. Time based check isn't reliable as queue might not
> run at given time, so uses a small time isn't ok.
1s is too much, but I wouldn't abandon a time based approach. To fix
the problem of queue not being run, you can consider a slice. If at
the end of the slice, the queue is seeky, you split it.

> I'm thinking changing the split
> check based on requests number instead of time. That is if several continuous
> requests are regarded as seeky, the coop queue is split. See blow RFC patch.
> How many count a queue should be split after need more consideration,
> below patch just uses an arbitary number.  This reduce about 5% performance
> lost when doing tio 32 threads sequential read.

Thanks,
Corrado
-- 
__________________________________________________________________________

dott. Corrado Zoccolo                          mailto:czoccolo@gmail.com
PhD - Department of Computer Science - University of Pisa, Italy
--------------------------------------------------------------------------