[Bioclusters] Tool to benchmark disk IO?

Joe Landman bioclusters@bioinformatics.org
Fri, 24 Sep 2004 13:58:42 -0400

Hi Dan:

  bonnie++ is good (http://www.coker.com.au/bonnie++/) as is iozone 
(http://www.iozone.org/).  Lots of folks have written tools to do this, 
but the important thing is to test your own workloads.  What apps are 
you running which are disk intensive?  This will indicate what the IO 
pattern (if any) looks like, and suggest the appropriate tests.  If you 
have the requisite program in place, it is trivial to create a bbsv3 
(http://bioinformatics.org/bbs) xml package to do the testing.

  As for doing things over NFS, there are positives and negatives to 
it.  Big negative is when you get lots of traffic.  If you need to work 
your way around that, there are a number of strategies that we could 
talk about.  Local IO is almost always fastest unless you use a system 
like Panasas.  There are other tradeoffs when you go these routes.  
SANs/NASs fall into the in-between area.


Dan Bolser wrote:

>Hi, I have checked the archive, but I can't find any reference to a simple
>IO 'tester' app to benchmark disk IO.
>I am fairly sure it would be easy to write a little tool to do this. Has
>anyone done that?
>I would like to be able to say things like
>ioRateTool --test-file /IDEDiskMount/testFile.scratch
>ioRateTool --test-file /SCSIDiskMount/testFile.scratch
>ioRateTool --test-file /NFSDiskMount/testFile.scratch
>ioRateTool --test-file /ramfsDiskMount/testFile.scratch
>ioRateTool --test-file /dev/null/testFile.scratch
>Although I guess that last one could only measure 'disk' write speed and
>not read speed.
>I want to see if my current NFS settings can be optimized, and I would
>like a clean (and hopefully robust) way to check what effect changing the
>options have.
>Thanks for any pointers,
>Ahhh... I just found the collective NFS rant... 
>We use NFS as our principal 'communication' channel, that is to say it is
>our group sand box. We put all our downloaded databases / scripts / data /
>homes on one big nfs server. This allows us to work on the same data
>easily, share files / results etc.
>Most of our 'day to day' computing is over this data. 
>How many Hail Mary's should I say?
>Bioclusters maillist  -  Bioclusters@bioinformatics.org

Joseph Landman, Ph.D
Scalable Informatics LLC,
email: landman@scalableinformatics.com
web  : http://scalableinformatics.com
phone: +1 734 612 4615