Re: [RFC V2 PATCH 5/5] cfq-iosched: fairness for sync no-idle queues

[Corrado Zoccolo <czoccolo@gmail.com>]

On Mon, Oct 26, 2009 at 4:45 PM, Jeff Moyer <jmoyer@redhat.com> wrote:
> Corrado Zoccolo <czoccolo@gmail.com> writes:
>
>> +static enum wl_type_t cfqq_type(struct cfq_queue *cfqq)
>> +{
>> +     if (!cfq_cfqq_sync(cfqq))
>> +             return ASYNC_WORKLOAD;
>> +     if (CFQQ_SEEKY(cfqq) || !cfq_cfqq_idle_window(cfqq))
>> +             return SYNC_NOIDLE_WORKLOAD;
>> +     return SYNC_WORKLOAD;
>> +}
>
> The calculation to determine whether or not to idle is now done in at
> least 2 places.  Perhaps it's time for a helper function.

Maybe that CFQQ_SEEKY(cfqq) test can just be removed, and we just rely
on the idle flag set by cfq_update_idle_window (it is already testing
seekiness there, but it also considers number of samples).
I'll do some testing without it to see if it changes anything.

>
> There are some other coding style nits, but I'll let Jens comment on
> those.  I think the idea looks good.  Once you rebase I'll run some
> tests; this is some really good work!
>
Thanks. Testing on different hardware is always appreciated.

Cheers,
Corrado
> Cheers,
> Jeff
>