Re: Re-4: verify seems not to take place on iSCSI-Target writing

[Jens Axboe <axboe@kernel.dk>]

On 03/05/2012 10:11 AM, Stefan Hübner wrote:
> -------- Original Message --------
> Subject: Re: Re-2: verify seems not to take place on iSCSI-Target writing (02-Mrz-2012 19:46)
> From:    Jens Axboe <axboe@kernel.dk>
> To:      shu@xtivate.de
> 
>> On 2012-03-02 18:06, Stefan Hübner wrote:
>>> Dear Jens,
>>>
>>> this is the output.  I added some supplementary details in the hope it's 
>>> useful ...
>>>
>>> UltraBitch ~ # ISCSI_DEVICE=/dev/sdb SECONDS=$((20*3600)) fio iscsi-stress.
>>> fio 
>>> iSCSI: (g=0): rw=randwrite, bs=4K-4M/4K-4M, ioengine=libaio, iodepth=16
>>> fio 2.0.3
>>> Starting 1 process
>>> Jobs: 1 (f=1): [w] [17.3% done] [0K/113.2M /s] [0 /71  iops] [eta 03d:23h:
>>> 28m:21s]
>>> iSCSI: (groupid=0, jobs=1): err= 0: pid=18184
>>>   write: io=7740.2GB, bw=112723KB/s, iops=63 , runt=72000222msec
>>>     bw (KB/s)  : min= 3669, max=593424, per=100.00%, avg=113561.25, stdev=
>>> 26868.27
>>>   cpu          : usr=4.18%, sys=1.28%, ctx=4760480, majf=0, minf=106912
>>>   IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=100.0%, 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=0.1%, 32=0.0%, 64=0.0%, >=64=
>>> 0.0%
>>>      issued    : total=r=0/w=4560578/d=0, short=r=0/w=0/d=0
>>>
>>> Run status group 0 (all jobs):
>>>   WRITE: io=7740.2GB, aggrb=112723KB/s, minb=115428KB/s, maxb=115428KB/s, 
>>> mint=72000222msec, maxt=72000222msec
>>
>> No reads happened - quick guess, you had the timeout hit before the
>> write phase has finished. Hence it'll exit, and don't do verifies.
>>
>> The reads you saw in the stats must be from other processes. Fio didn't
>> do any.
>>
>> -- 
>> Jens Axboe
>>
> Hi Jens,
> 
> then I don't quite understand what sense "verify_async" makes.  I
> thought it'd do the verifies "in parallel".
> 
> Any hint's how to get a timebased torture-test with verify working?

You didn't add a specific backlog for the verify threads. Granted that's
a bit confusing... But you want to tell fio how far behind the verify
threads are running. So if you set verify_backlog=32 for instance, it
will be 32 ios behind the writer. That will give you a job that
continually randomly writes and verifies as it goes, the kind of torture
I think you are looking for.

verify_async=2 just tells fio to allocate 2 threads for doing verifies.

-- 
Jens Axboe