Re: FIO creating empty latency log files

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 10/28/2015 12:41 AM, Buttetsu Batou wrote:
On 06/04/2015 05:09 PM, Reed Crowe wrote:
Using this version of FIO:
* io-threads
     master
rcrowe@rcrowe-desktop:~/fio_iothreads/fio$ ./fio --version
fio-2.2.7-14-g00b2

I have a job file which looks like this:

[global]
ioengine=libaio
direct=1
numjobs=1
bs=4k
iodepth=300
rw=randwrite
randrepeat=0
io_submit_mode=offload
rate_iops=8070
time_based=1
runtime=1m

[slow_write]
filename=/mnt/test1/testfile.io
write_lat_log=/tmp/fio_lat

And it produces empty files:
-rw-r--r-- 1 root   root         0 Jun  4 14:34 fio_lat_clat.1.log
-rw-r--r-- 1 root   root         0 Jun  4 14:34 fio_lat_lat.1.log
-rw-r--r-- 1 root   root         0 Jun  4 14:34 fio_lat_slat.1.log

What am I missing?

I am seeing this behavior in the latest release fio-2.2.10 and in
master fio-2.2.10-29-g8a68 but I am not using offload.

Is anyone currently able to successfully generate log output? If so
could you share your fio version and your config?

Here is the job file I am using at the moment.. have tried variations
on these settings and different fio versions back to 2.0.15 but always
the same empty output:

[write]
group_reporting=1
numjobs=8
ioengine=posixaio
io_submit_mode=inline
stonewall
rw=write
bs=32k
time_based=1
runtime=10
filesize=64k
directory=/tmp
per_job_logs=1
write_bw_log=write_bw
write_iops_log=write_iops

Updated the issue reported here:

https://github.com/axboe/fio/issues/29

Thank you for any ideas you may have!

I think your issue is because you set filesize=64k. Your job basically sets up 8 files of 64k each, then writes to them for 10 seconds. This means that the fio main loop keeps re-entering for every 64k for each thread. The bandwidth/iops is averaged over a window of 500msec by default, and the main loop runs for much shorter than that since it only does 64k each time. Hence you never get logged any entries.

Now, it should work. If you comment out these two lines:

memcpy(&td->bw_sample_time, &td->start, sizeof(td->start));
memcpy(&td->iops_sample_time, &td->start, sizeof(td->start));

in backend.c around 1568/1569, then I suspect the logging should improve for you. I'll have to see if we can safely do that without making other changes, before committing a change like that.t

--
Jens Axboe

--
To unsubscribe from this list: send the line "unsubscribe fio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html



[Index of Archives]     [Linux Kernel]     [Linux SCSI]     [Linux IDE]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux SCSI]

  Powered by Linux