[Bioclusters] Re: resources on administering clusters

[Joe Landman]

On Tue, 2002-04-02 at 14:33, James Cuff wrote:

[...]

> I guess my only question is whether anyone else used this type of software
> RAID 0 on IDE in production farms before and are there any real gotchas,
> tuning/tweaking or total nightmares I ought to be aware of?  I have enough
> nightmares, and would like to keep them to a minimum :-)

On the tuning side, make sure your chunksize is set appropriately in
/etc/raidtab

The default value is not optimized for large block sequential reads 
(e.g. streaming databases from disk).  

Also, see if your chipsets support reliable UDMA/DMA operation.  There
are many other things you can do to optimize IO.  Usually it is good
practice to run

	vmstat 1 > log

on a machine that you suspect of IO problems.  This will give you a log
snapshot of the high level system view.  If you fill up your IO channel,
or your disk is your limiting factor, you will be able to see it (with
some practice/coaching) from this log.


-- 
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