[Bioclusters] Tool to benchmark disk IO?

James Cuff bioclusters@bioinformatics.org
Fri, 24 Sep 2004 14:32:20 -0400


On Fri, 2004-09-24 at 14:21, Dan Bolser wrote:
> On Fri, 24 Sep 2004, elijah wright wrote:

> Seeker 1...Seeker 3...Seeker 2...start 'em...done...done...done...
>               -------Sequential Output-------- ---Sequential Input--
> --Random--
>               -Per Char- --Block--- -Rewrite-- -Per Char- --Block---
> --Seeks---
> Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
>           100  4240 25.9  5437  2.1  4780  2.4 19595 100.1 566942 99.7 2023.0 10.1
> 
> With default file size (104857600). The CPU is 'amazingly high' during
> reads.

I think your file size is to small - 100MB will fit in the UBC.  Once
written according to these figures you are seeing ca 600MB/s reads for
getchar ops.  The block ops must not be htting the buffer cache
correctly so go back down to 2MB/s.  This seems pretty poor even for
NFS.  Best bet is to wind up the file size you use.  I'm guessing your
host has 512MB of memory at least try to double the amount of memory to
use as a file size test.  If you don't make the file size bigger you
can't trust the read speed.  

I could be off with my numbers, I've been up to my eyeballs in Oracle
all morning, and have drunk way too much coffee..

Best regards,

J.

--
James Cuff, D. Phil.
Group Leader, Applied Production Systems
Broad Institute of MIT and Harvard. 320 Charles Street,
Cambridge, MA. 02141.  Tel: 617-252-1925  Fax: 617-258-0903