Qn on setting ioengine on Replay job

Qn on setting ioengine on Replay job

[Lakshmi]

Hi Jens,

  Is it possible to set the ioengine and iodepth on a replay job?
Would it have any effect on the replay jobs? I tried setting as below,
but didnt see any change. So wanted to make sure if what I am doing is
valid.

[global]
replay_no_stall=1
ioengine=posixaio
buffered=0
iodepth=6

[job1]
read_iolog=/home/file_3080658.bin

==============================
Please let me know

Thanks,
Lakshmi

Re: Qn on setting ioengine on Replay job

[Jens Axboe]

On 04/08/2014 10:39 AM, Lakshmi wrote:
> Hi Jens,
>
>    Is it possible to set the ioengine and iodepth on a replay job?
> Would it have any effect on the replay jobs? I tried setting as below,
> but didnt see any change. So wanted to make sure if what I am doing is
> valid.
>
> [global]
> replay_no_stall=1
> ioengine=posixaio
> buffered=0
> iodepth=6
>
> [job1]
> read_iolog=/home/file_3080658.bin

The ioengine will always work, the iodepth should work as well. blktrace 
loading doesn't actually set the iodepth, so it should run at depth 1 
unless you tell it otherwise. That should be fixed up, actually.

That said, you wont see higher depth than the job itself has done. The 
depth will be naturally limited by the events in that trace. However, if 
a lower depth is set, then that should be enforced (since fio is forced 
to wait on IO if it runs out of IO buffers).

-- 
Jens Axboe

Re: Qn on setting ioengine on Replay job

[Jens Axboe]

On 04/08/2014 01:14 PM, Jens Axboe wrote:
> On 04/08/2014 10:39 AM, Lakshmi wrote:
>> Hi Jens,
>>
>>    Is it possible to set the ioengine and iodepth on a replay job?
>> Would it have any effect on the replay jobs? I tried setting as below,
>> but didnt see any change. So wanted to make sure if what I am doing is
>> valid.
>>
>> [global]
>> replay_no_stall=1
>> ioengine=posixaio
>> buffered=0
>> iodepth=6
>>
>> [job1]
>> read_iolog=/home/file_3080658.bin
>
> The ioengine will always work, the iodepth should work as well. blktrace
> loading doesn't actually set the iodepth, so it should run at depth 1
> unless you tell it otherwise. That should be fixed up, actually.
>
> That said, you wont see higher depth than the job itself has done. The
> depth will be naturally limited by the events in that trace. However, if
> a lower depth is set, then that should be enforced (since fio is forced
> to wait on IO if it runs out of IO buffers).

Added a patch to probe the depth from the blktrace file. So should now 
set the detected depth, if it can.

-- 
Jens Axboe