[pnfs] ior / iozone benchmark

[Dean Hildebrand]

Hey All,

Here is a page I wrote before leaving citi that describes how I was 
using IOR and iozone.  It is a little out of date, and some things don't 
apply anymore, but it has the basics of how to run IOR.
http://www.citi.umich.edu/projects/asci/benchmarks.html

Updates:
a) The ior link is now: http://sourceforge.net/projects/ior-sio
b) iozone can do multiple clients as well using cluster mode (and it 
doesn't use mpi which makes it easier to install), but I find it a 
little buggy and clunky, and it definitely doesn't scale past 5 nodes or so.
c) With IOR, I used to use the clearcache program at the bottom of the 
above link to purge the page cache so reads go to disk, but I've found 
this is no longer necessary.  IOR now seems to do something to clear the 
page cache.
d) For IOR, the most important switches are
    -F - multiple files
    -C - reorder clients after write to ensure clients do not read the 
data they just wrote.
   -b  - size of file
    -t - transfer size
e) I also like mdtest (http://sourceforge.net/projects/mdtest/) for 
parallel metadata testing.

Dean