Urgent problem while using Fio to test random write performance

Urgent problem while using Fio to test random write performance

[Nan WANG]

hi all~
     I wanted to use fio to test random write performance, the command 
line is as forllowing:

       fio --filename=/dev/sdb1 --direct=1 --iodepth 1 --thread 
--rw=randwrite --ioengine=psync --bs=16k --size=200G --numjobs=10 
--runtime=1000 --group_reporting --name=mytest

     After the execution of fio finished, my disc usage become 100%, is 
there any method to release the content written by Fio?

  Thanks a lot in advance!!

-- 
王楠    技术研发中心
---------------------------------
==专业·责任·服务==
中国互联网络信息中心 CNNIC
网 址: www.cnnic.cn
中国互联网络信息中心.中国
地 址: 北京市海淀区中关村南4街4号
北京349信箱6分箱（100080）
---------------------------------
Nan WANG
Research&Development Center
= =Profession • Responsibility • Service= =
China Internet Network Information Center （CNNIC）
Website: www.cnnic.cn
          中国互联网络信息中心.中国
Add: 4th Zhongguancun Nansijie, Haidian District, Beijing 100190, China
POB: Beijing 349, Branch 6

Re: Urgent problem while using Fio to test random write performance

[David Nellans]

you're randwrite to a raw partition. if you had a filesystem on that 
partition you were trying to use, then well, your data is gone/corrupted.
without a filesystem on it, i'm not sure how you'd think that your disc 
usage has become "100%" without some xgui widgit warning you, so seems 
likely you did something you didn't mean to with regards to writing to 
/dev/sdb1 directly

if you use fio on devices/partitions with a filesystem, you need to have 
it write to files or it will trash your filesystem



On 05/21/2015 03:17 AM, Nan WANG wrote:
> hi all~
>     I wanted to use fio to test random write performance, the command 
> line is as forllowing:
>
>       fio --filename=/dev/sdb1 --direct=1 --iodepth 1 --thread 
> --rw=randwrite --ioengine=psync --bs=16k --size=200G --numjobs=10 
> --runtime=1000 --group_reporting --name=mytest
>
>     After the execution of fio finished, my disc usage become 100%, is 
> there any method to release the content written by Fio?
>
>  Thanks a lot in advance!!
>

Re: Urgent problem while using Fio to test random write performance

[Sitsofe Wheeler]

On 21 May 2015 at 15:22, David Nellans <david@nellans.org> wrote:
> you're randwrite to a raw partition. if you had a filesystem on that
> partition you were trying to use, then well, your data is gone/corrupted.
> without a filesystem on it, i'm not sure how you'd think that your disc
> usage has become "100%" without some xgui widgit warning you, so seems
> likely you did something you didn't mean to with regards to writing to
> /dev/sdb1 directly
>
> if you use fio on devices/partitions with a filesystem, you need to have it
> write to files or it will trash your filesystem

It's unlikely but it wasn't mentioned which partition became 100%
full. You should also make sure that /dev/sdb1 really exists as a
block device (e.g. via using file) otherwise you'll make/fill a
standard file called /dev/sdb1 .

-- 
Sitsofe | http://sucs.org/~sits/

Re: Urgent problem while using Fio to test random write performance

[Jens Axboe]

On 05/21/2015 02:39 PM, Sitsofe Wheeler wrote:
> On 21 May 2015 at 15:22, David Nellans <david@nellans.org> wrote:
>> you're randwrite to a raw partition. if you had a filesystem on that
>> partition you were trying to use, then well, your data is gone/corrupted.
>> without a filesystem on it, i'm not sure how you'd think that your disc
>> usage has become "100%" without some xgui widgit warning you, so seems
>> likely you did something you didn't mean to with regards to writing to
>> /dev/sdb1 directly
>>
>> if you use fio on devices/partitions with a filesystem, you need to have it
>> write to files or it will trash your filesystem
>
> It's unlikely but it wasn't mentioned which partition became 100%
> full. You should also make sure that /dev/sdb1 really exists as a
> block device (e.g. via using file) otherwise you'll make/fill a
> standard file called /dev/sdb1 .

That's true, though on modern systems that will fail since /dev is 
mounted on devtmpfs which doesn't support O_DIRECT. In general, 
allow_file_create=0 can be used to turn off file creation if you expect 
to run on block devices. This can avoid problems like this.


-- 
Jens Axboe