Re: [PATCH 1/1] dasd: do not force use of deadline IO scheduler

Re: [PATCH 1/1] dasd: do not force use of deadline IO scheduler

[Xose Vazquez Perez]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1860 bytes --]

On 2008-03-04 8:52:10, Martin Schwidefsky wrote:

a bit old.

> On Mon, 2008-03-03 at 15:21 -0500, Josef 'Jeff' Sipek wrote:
>> Signed-off-by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>
>> ---
>>  drivers/s390/block/dasd.c |    9 ---------
>>  1 files changed, 0 insertions(+), 9 deletions(-)
>> 
>> diff --git a/drivers/s390/block/dasd.c b/drivers/s390/block/dasd.c
>> index ccf46c9..4d4f6ff 100644
>> --- a/drivers/s390/block/dasd.c
>> +++ b/drivers/s390/block/dasd.c
>> @@ -1946,21 +1946,12 @@ static void do_dasd_request(struct request_queue *queue)
>>   */
>>  static int dasd_alloc_queue(struct dasd_block *block)
>>  {
>> -	int rc;
>> -
>>  	block->request_queue = blk_init_queue(do_dasd_request,
>>  					       &block->request_queue_lock);
>>  	if (block->request_queue == NULL)
>>  		return -ENOMEM;
>> 
>>  	block->request_queue->queuedata = block;
>> -
>> -	elevator_exit(block->request_queue->elevator);
>> -	rc = elevator_init(block->request_queue, "deadline");
>> -	if (rc) {
>> -		blk_cleanup_queue(block->request_queue);
>> -		return rc;
>> -	}
>>  	return 0;
>>  }
> 
> Um, why? We have a reason to force the deadline scheduler, our
> performance test have shown that the deadline scheduler get the highest
> throughout out of the storage subsystems which are used with the
> machines.

Are there recent tests, cfq vs. deadline, with current kernels?

This FCP/SCSI test is a bit old(2005-03-16), and cfq vs. deadline shows no differences:
http://www.ibm.com/developerworks/linux/linux390/perf/tuning_res_dasd_ioScheduler.html

Any DASD test out there ?

-thanks-

-- 
�All� muevan feroz guerra, ciegos reyes por un palmo m�s de tierra;
que yo aqu� tengo por m�o cuanto abarca el mar brav�o, a quien nadie
impuso leyes. Y no hay playa, sea cualquiera, ni bandera de esplendor,
que no sienta mi derecho y d� pecho a mi valor.�

Re: [PATCH 1/1] dasd: do not force use of deadline IO scheduler

[Heiko Carstens]

> > Considering the call trace you sent:
> > 
> > > -	elevator_exit(block->request_queue->elevator);
> > > -	rc = elevator_init(block->request_queue, "deadline");
> > 
> > block->request_queue->elevator = NULL is missing here.
> > 
> > > -	if (rc) {
> > > -		blk_cleanup_queue(block->request_queue);
> > > -		return rc;
> > > -	}
> > 
> > Fallback to noop and printing a warning wouldn't be bad.
> 
> Is noop better than the default (e.g., CFQ)?
> 
> I was thinking that if the deadline init fails, we put back the old elevator
> - whatever it may be.
> 
> > In addition something like depends on IOSCHED_DEADLINE should be added to
> > the Kconfig file.
> 
> Alright, so...
> 
> 1) Kconfig change, ->elevator = NULL

That should be the way to go, since we want the deadline scheduler.