[OpenAFS] Using 'fs storebehind'

[Penney Jr., Christopher (C.)]

I was experimenting with this tuning option and I can't seem to realize any
difference in performance.  I'm using 'iozone' to test performance on a
couple of Sun Solaris 9 boxes using OpenAFS 1.2.11 and I'm trying to
optimize for writing a large file into AFS space.  I currently have a 1GB
AFS cache and I'm writing a 128MB file into AFS space.  With 'fs storebehind
-allfiles 0' I get 21MB/s.  With 'fs storebehind -allfiles 131072' (128MB) I
get the same basic value.  

The command line I'm using is: iozone -c -i 0 -s 128m -r 32k -f testfile

Does this 'storebehind' option simply not work under Solaris?  I wonder
because when I use 'iostat -x -z 5' to watch disk performance the write rate
is very bursty.  I've see up to 40-60MB/s on the AFS cache LUN (it's a
stripe from a SAN), but then it trails off and then comes back.

On a side note, does performance of AFS cache deteriorate as the cache gets
larger?  If I wanted a 128GB AFS cache would I be in trouble?

Any advice?

   Chris