io_size usage

io_size usage

[Jun He]

Hi all,
I think there is a bug for io_size. 

For example, I want to write 16KB data to a file of size 8KB (so each block is
written twice) and I have the following job file:

[global]
ioengine=sync
rw=write
bs=4kb
size=8kb
io_size=32kb

[job1]
write_iolog=iolog.1


After running, iolog.1 will contain the following contents:

fio version 2 iolog
job1.0.0 add
job1.0.0 open
job1.0.0 write 0 4096
job1.0.0 write 4096 4096
job1.0.0 close

This shows that each block has been only written once. 

If I change the ioengine to libaio, iolog.1 will contain:

fio version 2 iolog
job1.0.0 add
job1.0.0 open
job1.0.0 write 0 4096
job1.0.0 write 4096 4096
job1.0.0 close
job1.0.0 open
job1.0.0 write 0 4096
job1.0.0 write 4096 4096
job1.0.0 close
job1.0.0 open
job1.0.0 write 0 4096
job1.0.0 write 4096 4096
job1.0.0 close
job1.0.0 open
job1.0.0 write 0 4096
job1.0.0 write 4096 4096
job1.0.0 close


Bug or wrong usage?

Thanks,
Jun

Re: io_size usage

[Sitsofe Wheeler]

Hi,

Your job sets and io_size four times that of "size" of the disk that
you are using. This will force your I/O to wrap abound to the start of
the region (why not use loops?). Are you saying that you think the
sync job's iolog is wrong because it only says it performed one loop?

On 9 December 2015 at 02:50, Jun He <jhe@cs.wisc.edu> wrote:
> Hi all,
> I think there is a bug for io_size.
>
> For example, I want to write 16KB data to a file of size 8KB (so each block is
> written twice) and I have the following job file:
>
> [global]
> ioengine=sync
> rw=write
> bs=4kb
> size=8kb
> io_size=32kb
>
> [job1]
> write_iolog=iolog.1
>
>
> After running, iolog.1 will contain the following contents:
>
> fio version 2 iolog
> job1.0.0 add
> job1.0.0 open
> job1.0.0 write 0 4096
> job1.0.0 write 4096 4096
> job1.0.0 close
>
> This shows that each block has been only written once.
>
> If I change the ioengine to libaio, iolog.1 will contain:
>
> fio version 2 iolog
> job1.0.0 add
> job1.0.0 open
> job1.0.0 write 0 4096
> job1.0.0 write 4096 4096
> job1.0.0 close
> job1.0.0 open
> job1.0.0 write 0 4096
> job1.0.0 write 4096 4096
> job1.0.0 close
> job1.0.0 open
> job1.0.0 write 0 4096
> job1.0.0 write 4096 4096
> job1.0.0 close
> job1.0.0 open
> job1.0.0 write 0 4096
> job1.0.0 write 4096 4096
> job1.0.0 close
>
>
> Bug or wrong usage?
>
> Thanks,
> Jun
>
>
>
> --
> To unsubscribe from this list: send the line "unsubscribe fio" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html



-- 
Sitsofe | http://sucs.org/~sits/

Re: io_size usage

[Jun He]

Thanks for the mentioning the 'loops' option.

But still, it is weird that by changing ioengine, iolog will report differently.
I think either sync mistakenly performs only one loop, or the IO is done
correctly but iolog report is wrong.

I think the expected behavior is that iolog is the same for both sync and
libaio. And both of them should follow the HOW:

"if 'size' is set to 21G, and 'io_size' is set to 40G, then fio will do 40G of
IO within the 0..20G region."


Jun




On Thu, Dec 10, 2015 at 12:56 AM, Sitsofe Wheeler <sitsofe@gmail.com> wrote:
> Hi,
>
> Your job sets and io_size four times that of "size" of the disk that
> you are using. This will force your I/O to wrap abound to the start of
> the region (why not use loops?). Are you saying that you think the
> sync job's iolog is wrong because it only says it performed one loop?
>
> On 9 December 2015 at 02:50, Jun He <jhe@cs.wisc.edu> wrote:
>> Hi all,
>> I think there is a bug for io_size.
>>
>> For example, I want to write 16KB data to a file of size 8KB (so each block is
>> written twice) and I have the following job file:
>>
>> [global]
>> ioengine=sync
>> rw=write
>> bs=4kb
>> size=8kb
>> io_size=32kb
>>
>> [job1]
>> write_iolog=iolog.1
>>
>>
>> After running, iolog.1 will contain the following contents:
>>
>> fio version 2 iolog
>> job1.0.0 add
>> job1.0.0 open
>> job1.0.0 write 0 4096
>> job1.0.0 write 4096 4096
>> job1.0.0 close
>>
>> This shows that each block has been only written once.
>>
>> If I change the ioengine to libaio, iolog.1 will contain:
>>
>> fio version 2 iolog
>> job1.0.0 add
>> job1.0.0 open
>> job1.0.0 write 0 4096
>> job1.0.0 write 4096 4096
>> job1.0.0 close
>> job1.0.0 open
>> job1.0.0 write 0 4096
>> job1.0.0 write 4096 4096
>> job1.0.0 close
>> job1.0.0 open
>> job1.0.0 write 0 4096
>> job1.0.0 write 4096 4096
>> job1.0.0 close
>> job1.0.0 open
>> job1.0.0 write 0 4096
>> job1.0.0 write 4096 4096
>> job1.0.0 close
>>
>>
>> Bug or wrong usage?
>>
>> Thanks,
>> Jun
>>
>>
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe fio" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
>
>
> --
> Sitsofe | http://sucs.org/~sits/