[ILUG] I/O benchmarking tools

[Conor Wynne]

On 16/02/2011 17:01, Brian O'Mahony wrote:
> Conor
>
> I would be interested in your findings also. 

Sure, but it will take a couple of weeks to finish.

> Regards
>
> B
> -----Original Message-----
> From: ilug-bounces at linux.ie [mailto:ilug-bounces at linux.ie] On Behalf Of Conor Wynne
> Sent: Wednesday, February 16, 2011 12:01 PM
> To: Pádraig Brady
> Cc: ilug at linux.ie
> Subject: Re: [ILUG] I/O benchmarking tools
>
> On 16/02/2011 11:18, Pádraig Brady wrote:
>> On 16/02/11 10:18, stephen mulcahy wrote:
>>> On 15/02/11 14:12, Conor Wynne wrote:
>>>> What tools do you guys recommend for doing some I/O benchmarking.
>>>>
>>>> I've been using iozone with 20 threads, but the figures I'm getting 
>>>> back are just mental. Like more than the array should be capable of....
>>>>
>>>> The is the comand line I've been testing with:
>>>>
>>>> /opt/iozone/bin/iozone -R -l 20 -u 20 -r 4k -s 100m -F file1 file2 
>>>> file3 file4 file5 file6 file7 file8 file9 file10
>>>> file11 file12 file13 file14 file15 file16 file17 file18 file19 
>>>> file20
>>>> |tee -a /root/Test_20-threads_Iozone.txt
>>> iozone has a pretty good reputation.
>>>
>>> I like bonnie++ because it has a simpler interface but afaik iozone 
>>> is more accurate and more powerful.
>>>
>>> if you're doing any filesystem testing, the first thing you need to 
>>> be careful of is testing the performance of your os caches rather 
>>> than the actual filesystem. So to be sure to be sure, you should use 
>>> a test-file size that is a multiple of your system's memory size 
>>> (this is certainly the case with bonnie++ and I assume the same with iozone).
>>>
>>> Try that and see if it gives any improvement.
>> Re caching, since Linux 2.6.16 one can drop all cache with:
>>
>> echo 1 > /proc/sys/vm/drop_caches # free pagecache echo 2 > 
>> /proc/sys/vm/drop_caches # free dentries and inodes echo 3 > 
>> /proc/sys/vm/drop_caches # free all of the above
>>
>> Note, "dirty" stuff is not freed, so one should run `sync` first.
>>
>> Coincidentally I was adding a per file cache drop to `dd` last night.
>> It will look something like:
>>
>> dd if=file iflag=nocache count=0 # drop all of file from cache dd 
>> if=file iflag=nocache skip=10 count=0 # drop end of file from cache dd 
>> if=file iflag=nocache skip=10 count=10 # drop part of file from cache 
>> dd if=file iflag=nocache of=copy oflag=nocache # copy just using 
>> readahead cache
>>
>> I'm wondering if I need to imply fdatasync with nocache.
>> Also perhaps conv=stream would be a useful shortcut (for the last 
>> example at least).
>>
>> Suggestions welcome.
>>
>> cheers,
>> Pádraig.
> Excellent stuff, thanks to you both. Now I can settle some long running arguments over which to use, mpio or bonding for iSCSI targets... :)
>
> --
> Irish Linux Users' Group mailing list
> About this list : http://mail.linux.ie/mailman/listinfo/ilug
> Who we are : http://www.linux.ie/
> Where we are : http://www.linux.ie/map/
>
>
> The information in this email is confidential and may be legally privileged.
> It is intended solely for the addressee. Access to this email by anyone else
> is unauthorized. If you are not the intended recipient, any disclosure,
> copying, distribution or any action taken or omitted to be taken in reliance
> on it, is prohibited and may be unlawful. If you are not the intended
> addressee please contact the sender and dispose of this e-mail. Thank you.
>