fio queue depth and windows build.

fio queue depth and windows build.

[Iuliu Rus]

Hello,
Looking at fio for windows, we are pretty sure that the queue depth
parameter is not working.
We wanted to investigate this ourselves and prepare a fix, but the
windows build is not working either. From a cygwin64 shell, make clean
and make -j results in this:
t/dedupe.c:13:23: fatal error: sys/ioctl.h: No such file or directory
(although i see ioctl.h under cygwin64\usr\include\sys and under
cygwin64\usr\i686-pc-cygwin\sys-root\usr\include\sys).
Any idea why the build fails?

Re: fio queue depth and windows build.

[Bruce Cran]

I've been meaning to investigate that build failure: it does, however get as far as creating the fio.exe binary so you can ignore the error.

-- 
Bruce

> On Apr 25, 2015, at 12:04, Iuliu Rus <iuliur@google.com> wrote:
> 
> Hello,
> Looking at fio for windows, we are pretty sure that the queue depth
> parameter is not working.
> We wanted to investigate this ourselves and prepare a fix, but the
> windows build is not working either. From a cygwin64 shell, make clean
> and make -j results in this:
> t/dedupe.c:13:23: fatal error: sys/ioctl.h: No such file or directory
> (although i see ioctl.h under cygwin64\usr\include\sys and under
> cygwin64\usr\i686-pc-cygwin\sys-root\usr\include\sys).
> Any idea why the build fails?
> --
> 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

Re: fio queue depth and windows build.

[Iuliu Rus]

It doesn't create fio.exe for me. I commented that include out since
it doesn't look like it's needed, but then linker complains about mmap
munmap and sysconfig.


On Sat, Apr 25, 2015 at 12:46 PM, Bruce Cran <bruce@cran.org.uk> wrote:
> I've been meaning to investigate that build failure: it does, however get as far as creating the fio.exe binary so you can ignore the error.
>
> --
> Bruce
>
>> On Apr 25, 2015, at 12:04, Iuliu Rus <iuliur@google.com> wrote:
>>
>> Hello,
>> Looking at fio for windows, we are pretty sure that the queue depth
>> parameter is not working.
>> We wanted to investigate this ourselves and prepare a fix, but the
>> windows build is not working either. From a cygwin64 shell, make clean
>> and make -j results in this:
>> t/dedupe.c:13:23: fatal error: sys/ioctl.h: No such file or directory
>> (although i see ioctl.h under cygwin64\usr\include\sys and under
>> cygwin64\usr\i686-pc-cygwin\sys-root\usr\include\sys).
>> Any idea why the build fails?
>> --
>> 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

Re: fio queue depth and windows build.

[Iuliu Rus]

It however did create the fio.exe after i commented out the include.
Thanks! We'll submit a patch once we figure out why the queue depth is
not working.
BTW this is Google Compute Engine team, and we're using fio to compare
our windows and linux storage performance.

On Sat, Apr 25, 2015 at 1:35 PM, Iuliu Rus <iuliur@google.com> wrote:
> It doesn't create fio.exe for me. I commented that include out since
> it doesn't look like it's needed, but then linker complains about mmap
> munmap and sysconfig.
>
>
> On Sat, Apr 25, 2015 at 12:46 PM, Bruce Cran <bruce@cran.org.uk> wrote:
>> I've been meaning to investigate that build failure: it does, however get as far as creating the fio.exe binary so you can ignore the error.
>>
>> --
>> Bruce
>>
>>> On Apr 25, 2015, at 12:04, Iuliu Rus <iuliur@google.com> wrote:
>>>
>>> Hello,
>>> Looking at fio for windows, we are pretty sure that the queue depth
>>> parameter is not working.
>>> We wanted to investigate this ourselves and prepare a fix, but the
>>> windows build is not working either. From a cygwin64 shell, make clean
>>> and make -j results in this:
>>> t/dedupe.c:13:23: fatal error: sys/ioctl.h: No such file or directory
>>> (although i see ioctl.h under cygwin64\usr\include\sys and under
>>> cygwin64\usr\i686-pc-cygwin\sys-root\usr\include\sys).
>>> Any idea why the build fails?
>>> --
>>> 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

Re: fio queue depth and windows build.

[Iuliu Rus]

FYI i have sent a pull request https://github.com/axboe/fio/pull/64
which makes fio on windows on par with iometer. The difference quite
big on our local flash machines (60k iops to 100k)

On Sat, Apr 25, 2015 at 1:43 PM, Iuliu Rus <iuliur@google.com> wrote:
> It however did create the fio.exe after i commented out the include.
> Thanks! We'll submit a patch once we figure out why the queue depth is
> not working.
> BTW this is Google Compute Engine team, and we're using fio to compare
> our windows and linux storage performance.
>
> On Sat, Apr 25, 2015 at 1:35 PM, Iuliu Rus <iuliur@google.com> wrote:
>> It doesn't create fio.exe for me. I commented that include out since
>> it doesn't look like it's needed, but then linker complains about mmap
>> munmap and sysconfig.
>>
>>
>> On Sat, Apr 25, 2015 at 12:46 PM, Bruce Cran <bruce@cran.org.uk> wrote:
>>> I've been meaning to investigate that build failure: it does, however get as far as creating the fio.exe binary so you can ignore the error.
>>>
>>> --
>>> Bruce
>>>
>>>> On Apr 25, 2015, at 12:04, Iuliu Rus <iuliur@google.com> wrote:
>>>>
>>>> Hello,
>>>> Looking at fio for windows, we are pretty sure that the queue depth
>>>> parameter is not working.
>>>> We wanted to investigate this ourselves and prepare a fix, but the
>>>> windows build is not working either. From a cygwin64 shell, make clean
>>>> and make -j results in this:
>>>> t/dedupe.c:13:23: fatal error: sys/ioctl.h: No such file or directory
>>>> (although i see ioctl.h under cygwin64\usr\include\sys and under
>>>> cygwin64\usr\i686-pc-cygwin\sys-root\usr\include\sys).
>>>> Any idea why the build fails?
>>>> --
>>>> 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

Re: fio queue depth and windows build.

[Jens Axboe]

On 05/06/2015 12:35 PM, Iuliu Rus wrote:
> FYI i have sent a pull request https://github.com/axboe/fio/pull/64
> which makes fio on windows on par with iometer. The difference quite
> big on our local flash machines (60k iops to 100k)

Thanks for fixing this up, it's been a pain point for quite a while!

-- 
Jens Axboe

Re: fio queue depth and windows build.

[Alireza Haghdoost]

Did you guys reproduce exact the same IOMeter IOPS with
FIO on Windows ?  Would you please provide more validation details ?

On Thu, May 7, 2015 at 9:46 AM, Jens Axboe <axboe@kernel.dk> wrote:
> On 05/06/2015 12:35 PM, Iuliu Rus wrote:
>>
>> FYI i have sent a pull request https://github.com/axboe/fio/pull/64
>> which makes fio on windows on par with iometer. The difference quite
>> big on our local flash machines (60k iops to 100k)
>
>
> Thanks for fixing this up, it's been a pain point for quite a while!
>
> --
> Jens Axboe
>
>
> --
> 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

Re: fio queue depth and windows build.

[Iuliu Rus]

Not sure what validation details you need...
C:\>C:\fio-2.1.12-x64\fio --name=benchmark --rw=randread --ioengine=windowsaio -
-direct=1 --time_based --ramp_time=10 --runtime=90 --thread --numjobs=1 --invali
date=1    --filename=\\.\PhysicalDrive1 --nrfiles=1 --size=16G --bs=4k --iodepth
=128 --create_on_open=0 --norandommap --sync=0 --end_fsync=0 --randrepeat=1 --di
sable_lat=1 --disable_clat=1 --disable_slat=1 --disable_bw_measurement=1
benchmark: (g=0): rw=randread, bs=4K-4K/4K-4K/4K-4K, ioengine=windowsaio, iodept
h=128
fio-2.2.6
Starting 1 thread
Jobs: 1 (f=1): [r(1)] [11.0% done] [3155KB/0KB/0KB /s] [788/0/0 iops] [eta 01m:2
Jobs: 1 (f=1): [r(1)] [12.0% done] [203.2MB/0KB/0KB /s] [52.2K/0/0 iops] [eta 01
Jobs: 1 (f=1): [r(1)] [13.0% done] [211.6MB/0KB/0KB /s] [54.2K/0/0 iops] [eta 01
Jobs: 1 (f=1): [r(1)] [14.0% done] [210.7MB/0KB/0KB /s] [53.1K/0/0 iops] [eta 01
Jobs: 1 (f=1): [r(1)] [15.0% done] [218.1MB/0KB/0KB /s] [56.5K/0/0 iops] [eta 01
Jobs: 1 (f=1): [r(1)] [16.0% done] [221.1MB/0KB/0KB /s] [56.6K/0/0 iops] [eta 01

C:\>C:\my_fio\fio --name=benchmark --rw=randread --ioengine=windowsaio --direct=
1 --time_based --ramp_time=10 --runtime=90 --thread --numjobs=1 --invalidate=1
  --filename=\\.\PhysicalDrive1 --nrfiles=1 --size=16G --bs=4k --iodepth=128 --c
reate_on_open=0 --norandommap --sync=0 --end_fsync=0 --randrepeat=1 --clocksourc
e=cpu
fio: clocksource=cpu may not be reliable
benchmark: (g=0): rw=randread, bs=4K-4K/4K-4K/4K-4K, ioengine=windowsaio, iodept
h=128
fio-2.2.7-26-g77e7b
Starting 1 thread
Jobs: 1 (f=1): [r(1)] [11.0% done] [8977KB/0KB/0KB /s] [2244/0/0 iops] [eta 01m:
Jobs: 1 (f=1): [r(1)] [12.0% done] [366.2MB/0KB/0KB /s] [93.1K/0/0 iops] [eta 01
Jobs: 1 (f=1): [r(1)] [13.0% done] [352.5MB/0KB/0KB /s] [90.3K/0/0 iops] [eta 01
Jobs: 1 (f=1): [r(1)] [14.0% done] [373.1MB/0KB/0KB /s] [95.8K/0/0 iops] [eta 01

On Thu, May 7, 2015 at 8:05 AM, Alireza Haghdoost <haghdoost@gmail.com> wrote:
> Did you guys reproduce exact the same IOMeter IOPS with
> FIO on Windows ?  Would you please provide more validation details ?
>
> On Thu, May 7, 2015 at 9:46 AM, Jens Axboe <axboe@kernel.dk> wrote:
>> On 05/06/2015 12:35 PM, Iuliu Rus wrote:
>>>
>>> FYI i have sent a pull request https://github.com/axboe/fio/pull/64
>>> which makes fio on windows on par with iometer. The difference quite
>>> big on our local flash machines (60k iops to 100k)
>>
>>
>> Thanks for fixing this up, it's been a pain point for quite a while!
>>
>> --
>> Jens Axboe
>>
>>
>> --
>> 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

Re: fio queue depth and windows build.

[Bruce Cran]

I've just uploaded fio-2.2.7+ Windows installers and zip files that have 
this performance fix.  See http://bluestop.org/fio/ .  The '+' just 
means I didn't build against the tagged 2.2.7 sources.

-- 
Bruce

On 5/7/2015 11:10 AM, Iuliu Rus wrote:
> Not sure what validation details you need...
> C:\>C:\fio-2.1.12-x64\fio --name=benchmark --rw=randread --ioengine=windowsaio -
> -direct=1 --time_based --ramp_time=10 --runtime=90 --thread --numjobs=1 --invali
> date=1    --filename=\\.\PhysicalDrive1 --nrfiles=1 --size=16G --bs=4k --iodepth
> =128 --create_on_open=0 --norandommap --sync=0 --end_fsync=0 --randrepeat=1 --di
> sable_lat=1 --disable_clat=1 --disable_slat=1 --disable_bw_measurement=1
> benchmark: (g=0): rw=randread, bs=4K-4K/4K-4K/4K-4K, ioengine=windowsaio, iodept
> h=128
> fio-2.2.6
> Starting 1 thread
> Jobs: 1 (f=1): [r(1)] [11.0% done] [3155KB/0KB/0KB /s] [788/0/0 iops] [eta 01m:2
> Jobs: 1 (f=1): [r(1)] [12.0% done] [203.2MB/0KB/0KB /s] [52.2K/0/0 iops] [eta 01
> Jobs: 1 (f=1): [r(1)] [13.0% done] [211.6MB/0KB/0KB /s] [54.2K/0/0 iops] [eta 01
> Jobs: 1 (f=1): [r(1)] [14.0% done] [210.7MB/0KB/0KB /s] [53.1K/0/0 iops] [eta 01
> Jobs: 1 (f=1): [r(1)] [15.0% done] [218.1MB/0KB/0KB /s] [56.5K/0/0 iops] [eta 01
> Jobs: 1 (f=1): [r(1)] [16.0% done] [221.1MB/0KB/0KB /s] [56.6K/0/0 iops] [eta 01
>
> C:\>C:\my_fio\fio --name=benchmark --rw=randread --ioengine=windowsaio --direct=
> 1 --time_based --ramp_time=10 --runtime=90 --thread --numjobs=1 --invalidate=1
>    --filename=\\.\PhysicalDrive1 --nrfiles=1 --size=16G --bs=4k --iodepth=128 --c
> reate_on_open=0 --norandommap --sync=0 --end_fsync=0 --randrepeat=1 --clocksourc
> e=cpu
> fio: clocksource=cpu may not be reliable
> benchmark: (g=0): rw=randread, bs=4K-4K/4K-4K/4K-4K, ioengine=windowsaio, iodept
> h=128
> fio-2.2.7-26-g77e7b
> Starting 1 thread
> Jobs: 1 (f=1): [r(1)] [11.0% done] [8977KB/0KB/0KB /s] [2244/0/0 iops] [eta 01m:
> Jobs: 1 (f=1): [r(1)] [12.0% done] [366.2MB/0KB/0KB /s] [93.1K/0/0 iops] [eta 01
> Jobs: 1 (f=1): [r(1)] [13.0% done] [352.5MB/0KB/0KB /s] [90.3K/0/0 iops] [eta 01
> Jobs: 1 (f=1): [r(1)] [14.0% done] [373.1MB/0KB/0KB /s] [95.8K/0/0 iops] [eta 01
>
> On Thu, May 7, 2015 at 8:05 AM, Alireza Haghdoost <haghdoost@gmail.com> wrote:
>> Did you guys reproduce exact the same IOMeter IOPS with
>> FIO on Windows ?  Would you please provide more validation details ?
>>
>> On Thu, May 7, 2015 at 9:46 AM, Jens Axboe <axboe@kernel.dk> wrote:
>>> On 05/06/2015 12:35 PM, Iuliu Rus wrote:
>>>> FYI i have sent a pull request https://github.com/axboe/fio/pull/64
>>>> which makes fio on windows on par with iometer. The difference quite
>>>> big on our local flash machines (60k iops to 100k)
>>>
>>> Thanks for fixing this up, it's been a pain point for quite a while!
>>>
>>> --
>>> Jens Axboe
>>>
>>>
>>> --
>>> 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