Re: seems to be a flaw in cfq [message #9181 is a reply to message #9165] |
Thu, 21 December 2006 09:03 |
Vasily Tarasov
Messages: 1345 Registered: January 2006
|
Senior Member |
|
|
Hello, Jens,
Sorry for late answer.
The situation is the following:
direct seqread
(2.6.18) - reproducable
(2.6.20-rc1) - reproducable
(2.6.20-rc1 + your fix) - not reproducable
buffered randread
(2.6.18) - reproducable
(2.6.20-rc1) - not reproducable
(2.6.20-rc1 + your fix) - not reproducable
So the conclusion: problem with "direct seqread" is fixed by your patch
(thanks!) and
the problem with "buffered randread" was fixed in 2.6.19/2.6.20.
Thanks for cooperation,
Vasily.
Jens Axboe wrote:
> On Wed, Dec 20 2006, Jens Axboe wrote:
>
>> I'm still seeing odd results with buffered io, while the direct case is
>> definitely fixed. So we are either looking at two bugs, or one bug that
>> the merge fixup hides.
>>
>
> Ah hang on, buffered reads will be serialized by the page cache. If I
> rerun the buffered randread case with anticipatory, the results are
> skewed in the other direction but still unfair. So I'm not sure there's
> a real bug there in the io scheduler.
|
|
|