[nmglug] filesystem optimization on software raid 5

[Nick Frost]

Jason Schaefer wrote:

> So, back to "filesystem optimization" ??

Well, one approach would be to do some testing (if you have the time).

For example, you could use an OS drive and/ or a mirrored pair, and have 
the RAID-5 array for data storage.  Were it me, I might partition and 
format the array alternately as ext3, reiserfs, and XFS, and then run 
benchmarks with Bonnie++ and/or Iozone;

http://lbs.sourceforge.net/

Reiserfs is likely to be faster with smaller files, however, since Hans 
Reiser is in legal trouble, the status of the Reiserfs project is in 
question.

Ext3 is tried and true, but not fast.

You may find these threads helpful;

http://lists.freebsd.org/pipermail/freebsd-questions/2003-October/023669.html

http://lists.freebsd.org/pipermail/freebsd-performance/2004-February/000671.html

http://archives.postgresql.org/pgsql-performance/2006-07/msg00164.php

Basically if you want single/sequential I/O a smaller stripe size is 
better, but if you have multiple concurrent disk operations (several 
clients) a larger stripe size is better.

-Nick

-- 
           _      __   ____
    ____  (_)____/ /__/ __/
   / __ \/ / ___/ //_/ /_
  / / / / / /__/ ,< / __/
/_/ /_/_/\___/_/|_/_/
Nicholas S. Frost
nickf at nickorama.com