Re: question on sg ioengine

[Jens Axboe <axboe@kernel.dk>]

On 11/19/2014 03:50 PM, Zhe Zhang wrote:
> Hi,
>
> I run into this problem probably because I did not understand the
> command correctly.
>
> But my test.fio configured as follow:
> [global]
> bs=4k
> ba=512b
> direct=1
> number_ios=400
> rw=write:2k
> ioengine=sg
> iodepth=2
> iodepth_batch=2
> write_bw_log=test
> write_lat_log=test
> write_iops_log=test
> [/dev/sdb]
>
> and the result looks like this:
> /dev/sdb: (groupid=0, jobs=1): err= 0: pid=7783: Mon Nov 17 15:26:30 2014
>    write: io=1600.0KB, bw=6584.4KB/s, iops=1646, runt=   243msec
>      clat (usec): min=153, max=14926, avg=602.43, stdev=1165.02
>       lat (usec): min=153, max=14926, avg=602.91, stdev=1165.03
>      clat percentiles (usec):
>       |  1.00th=[  163],  5.00th=[  193], 10.00th=[  253], 20.00th=[  290],
>       | 30.00th=[  322], 40.00th=[  354], 50.00th=[  394], 60.00th=[  446],
>       | 70.00th=[  482], 80.00th=[  532], 90.00th=[  708], 95.00th=[ 1048],
>       | 99.00th=[ 6624], 99.50th=[ 7776], 99.90th=[14912], 99.95th=[14912],
>       | 99.99th=[14912]
>      lat (usec) : 250=9.25%, 500=66.00%, 750=15.00%, 1000=3.50%
>      lat (msec) : 2=3.50%, 4=0.75%, 10=1.75%, 20=0.25%
>    cpu          : usr=1.65%, sys=1.65%, ctx=402, majf=0, minf=36
>    IO depths    : 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0%
>       submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
>       complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
>       issued    : total=r=0/w=400/d=0, short=r=0/w=0/d=0
>       latency   : target=0, window=0, percentile=100.00%, depth=2
>
> Run status group 0 (all jobs):
>    WRITE: io=1600KB, aggrb=6584KB/s, minb=6584KB/s, maxb=6584KB/s,
> mint=243msec, maxt=243msec
>
> Disk stats (read/write):
>    sdb: ios=84/0, merge=0/0, ticks=92/0, in_queue=92, util=36.95%
>
> RUN FINISHED; exit value 0; real time: 600ms; user: 40ms; system: 60ms
>
> I darker my disk utilization here. My question is why there is no
> write went tot the disk and where does those 84 reads come from?
> Because in test.fio, I only issue writes to the disk /dev/sdb

Only file system requests add to the disk utilization stats in Linux. 
When you do sg based IO, those are not file system requests. So none of 
the writes done by fio would show up there.


-- 
Jens Axboe