topic Re: Calculating IOPS from bonnie++ results in Splunk Search

Calculating IOPS from bonnie++ results

ofrachon — Thu, 14 Feb 2013 17:16:07 GMT

Hello,

I was wondering how to obtain IOPS from bonnie++ results.

The various executions of bonnie++ have been done using a size of twice the RAM, and not the root user.

Here are the results:

# bonnie++ -d /data -s 12g -qfb
1.97,1.97,host28,1,1360870223,12G,,,,434320,86,116822,22,,,116187,9,1362,32,16,,,,,15457,27,+++++,+++,2854,9,32642,45,+++++,+++,1624,6,,212ms,158ms,,178ms,818ms,125ms,1310us,234ms,103ms,53424us,182ms

# bonnie++ -d /data -s 16g -qfb
1.97,1.97,host56,1,1360869078,16G,,,,392047,38,133174,5,,,363574,5,586.2,8,16,,,,,+++++,+++,+++++,+++,2679,1,+++++,+++,+++++,+++,2700,1,,226ms,289ms,,58929us,276ms,118us,76us,61321us,198us,77us,21293us

# bonnie++ -d /data -s 32g -qfb
1.97,1.97,host57,1,1360852635,32G,,,,505977,54,151226,3,,,254450,3,257.0,8,16,,,,,+++++,+++,+++++,+++,2748,1,+++++,+++,+++++,+++,2602,1,,226ms,438ms,,373ms,1261ms,217us,84us,107ms,131us,31us,65746us

# bonnie++ -d /data -s 16g -qfb
 1.97,1.97,host58,1,1360881644,16G,,,,478704,47,129047,4,,,330770,4,480.6,7,16,,,,,+++++,+++,+++++,+++,2195,2,+++++,+++,+++++,+++,2814,2,,373ms,389ms,,79935us,316ms,452us,82us,295ms,94us,169us,35213us

# bonnie++ -d /data -s 16g -qfb
1.97,1.97,host59,1,1360868560,16G,,,,396657,37,82812,1,,,246492,2,2800,43,16,,,,,+++++,+++,+++++,+++,2819,3,+++++,+++,+++++,+++,2761,1,,226ms,350ms,,90085us,70123us,582us,92us,61752us,135us,36us,13827us

Any help would be appreciated, as I just cannot find a way to translate those into the famed IOPS I am looking for.

The main idea here is to be able to find out if the 800 IOPS threshold is there or not on our future Splunk cluster.

Cheers,
Olivier

Re: Calculating IOPS from bonnie++ results

Ron_Naken — Thu, 14 Feb 2013 23:43:30 GMT

Be sure to use an updated version of bonnie++:
http://www.coker.com.au/bonnie++/

If you're using a version of bonnie that ships with the OS, you will likely find odd results. "Random Seeks/s" is the metric you want to measure, in order to estimate IOPS. Also be sure to test with -s set to 3x - 10x (ideally) memory to ensure no caching.

Re: Calculating IOPS from bonnie++ results

etbe — Tue, 19 Feb 2013 02:49:54 GMT

The default is to have the file size (the -s option) be twice the size of RAM which will usually be enough to break all caches. But if you need to test other things (like filesystem performance with big files) then feel free to make it as large as you wish.

When you use the -q option stdout gets the CSV format (as shown in the above question) and the human readable version goes to stderr. To convert the CSV version to human-readable use bon_csv2txt.

In the human readable form what you call "IOPS" is called "Random Seeks".

Re: Calculating IOPS from bonnie++ results

ivog — Sat, 28 Sep 2013 16:14:25 GMT

Hi,

is the measure for "Random Seeks" a representation of "Read IOPS", "Write IOPS", or some kind of average of both?

Thanks,
--ivo

Re: Calculating IOPS from bonnie++ results

ofrachon — Mon, 30 Sep 2013 09:21:05 GMT

It is some combined score of a sort, yes 🙂

Re: Calculating IOPS from bonnie++ results

michael_peters — Wed, 09 Apr 2014 15:15:14 GMT

Note that the IOPS requirement for SPLUNK 6.0 is 1200

The reason to use bonnie++ for comparability with what SPLUNK does ie repeatability of results. There are better tools to use

Re: Calculating IOPS from bonnie++ results

I_am_Jeff — Wed, 18 Jun 2014 16:07:49 GMT

Please share the "better tools" with links, please. (Not wanting to start a flame war. I'm interested in other tools.)

Re: Calculating IOPS from bonnie++ results

I_am_Jeff — Wed, 18 Jun 2014 16:10:21 GMT

+1 for telling those of us that have no clue, 'In the human readable form what you call "IOPS" is called "Random Seeks".'

Re: Calculating IOPS from bonnie++ results

prasadis — Fri, 30 Jan 2015 02:21:58 GMT

What you want to do is pass that input into bon_csv2txt. You should see a column named Random Seeks.

Re: Calculating IOPS from bonnie++ results

amrit — Thu, 10 Sep 2015 15:24:07 GMT

Per etbe's comment below, setting ram size to 3x-10x should be unnecessary.