runtime option not working similar for all devices

runtime option not working similar for all devices

[Vikram Seth]

Hi,

I am testing random reads on raw devices (hdd and flash) in a host. I
set same runtime value (12 hrs) in test file for each device.
However, while the test is running the ETA for hdd test (11:53 hrs) is
shown as more than twice of flash (4:40hrs). Later the test for flash
completes while the hdd test is still running.

I thought that having same runtime will cause the test to run for same
amount of time on all devices.
Is that not true ?

Thanks,
Vikram.

Re: runtime option not working similar for all devices

[Jens Axboe]

On 2011-09-12 06:33, Vikram Seth wrote:
> Hi,
> 
> I am testing random reads on raw devices (hdd and flash) in a host. I
> set same runtime value (12 hrs) in test file for each device.
> However, while the test is running the ETA for hdd test (11:53 hrs) is
> shown as more than twice of flash (4:40hrs). Later the test for flash
> completes while the hdd test is still running.
> 
> I thought that having same runtime will cause the test to run for same
> amount of time on all devices.
> Is that not true ?

The runtime option is a runtime cap, meaning that it'll stop the job
after the specified time only. If the job finishes before that (eg you
tell it to do 100g of IO and that takes a shorter time than the runtime
setting), then it'll exit normally.

-- 
Jens Axboe

Re: runtime option not working similar for all devices

[Vikram Seth]

On Mon, Sep 12, 2011 at 1:37 AM, Jens Axboe <axboe@kernel.dk> wrote:
> On 2011-09-12 06:33, Vikram Seth wrote:
>> Hi,
>>
>> I am testing random reads on raw devices (hdd and flash) in a host. I
>> set same runtime value (12 hrs) in test file for each device.
>> However, while the test is running the ETA for hdd test (11:53 hrs) is
>> shown as more than twice of flash (4:40hrs). Later the test for flash
>> completes while the hdd test is still running.
>>
>> I thought that having same runtime will cause the test to run for same
>> amount of time on all devices.
>> Is that not true ?
>
> The runtime option is a runtime cap, meaning that it'll stop the job
> after the specified time only. If the job finishes before that (eg you
> tell it to do 100g of IO and that takes a shorter time than the runtime
> setting), then it'll exit normally.
>
> --
> Jens Axboe
>
>

I did not give any other stop criteria, like io qty. The goal was to
stress the devices for a given amount of time.
Here is the test file for the disk (same for flash, except the filename option).

[global]
description=Standard Performance test
runtime=43200
direct=1
filename=/dev/sdc
numjobs=20
group_reporting

[test]
rw=randread
bs=128k

Thanks
Vikram.

Re: runtime option not working similar for all devices

[Josh Aune]

[-- Attachment #1: Type: text/plain, Size: 270 bytes --]

>
> [global]
> description=Standard Performance test
> runtime=43200
> direct=1
> filename=/dev/sdc
> numjobs=20
> group_reporting

Add time_based and it will run for the full duration specified by runtime.

Josh

>
> [test]
> rw=randread
> bs=128k
>
> Thanks
> Vikram.

[-- Attachment #2: Type: text/html, Size: 384 bytes --]

Re: runtime option not working similar for all devices

[Jens Axboe]

On 2011-09-13 10:04, Josh Aune wrote:
> 
>>
>> [global]
>> description=Standard Performance test
>> runtime=43200
>> direct=1
>> filename=/dev/sdc
>> numjobs=20
>> group_reporting
> 
> Add time_based and it will run for the full duration specified by runtime.

Yes. With the job given, once fio has done 128kb randreads to any block
on the total size of sdc, it will exit otherwise. If you only give
runtime=xx, then it's just a cap, it wont extend your runtime beyond the
size of the job.

-- 
Jens Axboe