Capturing idle time with blktrace

Capturing idle time with blktrace

[Brandon Schwartz]

Is there an accurate way to capture idle time with blktrace?  For
example, if I run/trace the following FIO script:

[global]
direct=1
ioengine=libaio
readwrite=read
blocksizeQ2k
filename=/dev/sdd

[read]
runtime\x10
name=read

[read2]
stonewall=1
startdelay@
runtime\x10
name=read2

And attempt to replay it using blkreplay, it will ignore the wait time
and simply read for 20seconds instead of read 10s, wait for 40s, read
10s.

Regards,
Brandon

Re: Capturing idle time with blktrace

[Brandon Schwartz]

And by blkreplay, I meant replay the blktrace with FIO

On Wed, Apr 22, 2015 at 2:52 PM, Brandon Schwartz <schwartz.xn@gmail.com> wrote:
> Is there an accurate way to capture idle time with blktrace?  For
> example, if I run/trace the following FIO script:
>
> [global]
> direct=1
> ioengine=libaio
> readwrite=read
> blocksizeQ2k
> filename=/dev/sdd
>
> [read]
> runtime\x10
> name=read
>
> [read2]
> stonewall=1
> startdelay@
> runtime\x10
> name=read2
>
> And attempt to replay it using blkreplay, it will ignore the wait time
> and simply read for 20seconds instead of read 10s, wait for 40s, read
> 10s.
>
> Regards,
> Brandon

Re: Capturing idle time with blktrace

[Jens Axboe]

On 04/22/2015 02:53 PM, Brandon Schwartz wrote:
> And by blkreplay, I meant replay the blktrace with FIO
>
> On Wed, Apr 22, 2015 at 2:52 PM, Brandon Schwartz <schwartz.xn@gmail.com> wrote:
>> Is there an accurate way to capture idle time with blktrace?  For
>> example, if I run/trace the following FIO script:
>>
>> [global]
>> direct=1
>> ioengine=libaio
>> readwrite=read
>> blocksizeQ2k
>> filename=/dev/sdd
>>
>> [read]
>> runtime\x10
>> name=read
>>
>> [read2]
>> stonewall=1
>> startdelay@
>> runtime\x10
>> name=read2
>>
>> And attempt to replay it using blkreplay, it will ignore the wait time
>> and simply read for 20seconds instead of read 10s, wait for 40s, read
>> 10s.


replay_no_stall=1

-- 
Jens Axboe

Re: Capturing idle time with blktrace

[Brandon Schwartz]

Ah, I think it was an issue with the version of FIO I was using.
Running 2.1.4 did not replay the blktrace properly and would ignore
the idle time between jobs.  I pulled 2.2.7 and it seems to be working
as expected, accurately including the idle time.

Regards,
Brandon

On Wed, Apr 22, 2015 at 4:51 PM, Jens Axboe <axboe@kernel.dk> wrote:
> On 04/22/2015 02:53 PM, Brandon Schwartz wrote:
>>
>> And by blkreplay, I meant replay the blktrace with FIO
>>
>> On Wed, Apr 22, 2015 at 2:52 PM, Brandon Schwartz <schwartz.xn@gmail.com>
>> wrote:
>>>
>>> Is there an accurate way to capture idle time with blktrace?  For
>>> example, if I run/trace the following FIO script:
>>>
>>> [global]
>>> direct=1
>>> ioengine=libaio
>>> readwrite=read
>>> blocksizeQ2k
>>> filename=/dev/sdd
>>>
>>> [read]
>>> runtime\x10
>>> name=read
>>>
>>> [read2]
>>> stonewall=1
>>> startdelay@
>>> runtime\x10
>>> name=read2
>>>
>>> And attempt to replay it using blkreplay, it will ignore the wait time
>>> and simply read for 20seconds instead of read 10s, wait for 40s, read
>>> 10s.
>
>
>
> replay_no_stall=1
>
> --
> Jens Axboe
>

Re: Capturing idle time with blktrace

[Alireza Haghdoost]

It should be capable to do that. Report your issue if you find out it
does not do the job. Capture your fio workload with blktrace, then run
captured traces through btrecord with small -m value and finally
replay back with btreplay.... Not sure how replay is useful in this
case when you do have access to fio to generate the same IO workload
for you.

On Wed, Apr 22, 2015 at 3:53 PM, Brandon Schwartz <schwartz.xn@gmail.com> wrote:
> And by blkreplay, I meant replay the blktrace with FIO
>
> On Wed, Apr 22, 2015 at 2:52 PM, Brandon Schwartz <schwartz.xn@gmail.com> wrote:
>> Is there an accurate way to capture idle time with blktrace?  For
>> example, if I run/trace the following FIO script:
>>
>> [global]
>> direct=1
>> ioengine=libaio
>> readwrite=read
>> blocksizeQ2k
>> filename=/dev/sdd
>>
>> [read]
>> runtime\x10
>> name=read
>>
>> [read2]
>> stonewall=1
>> startdelay@
>> runtime\x10
>> name=read2
>>
>> And attempt to replay it using blkreplay, it will ignore the wait time
>> and simply read for 20seconds instead of read 10s, wait for 40s, read
>> 10s.
>>
>> Regards,
>> Brandon
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrace" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html