[Bioclusters] Re: file server for cluster

Joe Landman bioclusters@bioinformatics.org
23 Apr 2002 15:16:23 -0400

On Tue, 2002-04-23 at 13:57, Ivo Grosse wrote:
> Hi Joe,
> Joe Landman <joe.landman@mscsoftware.com> wrote on Tue, 23 Apr 2002:
> > http://www.dell.com/us/en/esg/topics/segtopic_storage_scsi_main.htm you
> > see the 160 MB/s throughput.  This corresponds to 1280 Mb/s.
> You mean even if that disk array contained only 1 disk, it would give a 
> w/r flux of 160 MB/s?

No.  The disk itself (depending upon make, model, etc) can talk anywhere
from 10->50 MB/s.  The bus it is sitting on can deal with a maximum of
160 MB/s.  So you can (theoretically) fill up your 160 MB/s bus.  

> Naively, I was under the impression that only because of striping that 
> Gb/s flux could be achieved, but you say that a single disk can w/r 
> with 160 MB/s?

Striping is needed.  And this will work for specific IO patterns. 
Random I/O (reads and writes) on small blocks is usually the
pathological case for a fast file system hardware/software design. 
Large block sequential sustained reads are the optimal case.  The folks
doing database testing like to talk about disks in terms of Op/s, the
number of operations per second.


Joseph Landman, Ph.D.
Senior Scientist,
MSC Software High Performance Computing
email		: joe.landman@mscsoftware.com
messaging	: page_joe@mschpc.dtw.macsch.com
Main office	: +1 248 208 3312
Cell phone	: +1 734 612 4615
Fax		: +1 714 784 3774