ZFS Performance FreeBSD 9.0 vs. Openindiana

Bob Friesenhahn bfriesen at simple.dallas.tx.us
Thu Mar 29 14:46:29 UTC 2012


On Thu, 29 Mar 2012, Franz Schober wrote:
>
> For performance tests, I use the iozone benchmark with a multi-streaming 
> concurrency test,
> iozone -o -c -t 8 -r 128k -s 4G (Sync Mode, 8 concurrent workers, 128 k 
> Recordsize, 4G working file for every worker to run not only in cache).

The 128k record size is special.  I think that this is the transition 
point where zfs writes sync data directly to the pool disks rather 
than to the log disks.  I don't know which way it goes (pool/log) at 
exactly 128k.  If your log disks are idle during the benchmark, then 
the answer to this question would be clear.

> Now my question: Which parameters in the ZFS Subsystem of FreeBSD are 
> tuneable to reach the same performance in FreeBSD,
> especially pushing the synchronous write performance ?

I do not see much difference with the writes.  I do see a large 
difference with the reads.  This could easily be due to a difference 
in how Solaris and FreeBSD manage kernel memory.  In Solaris, 
substantial caching may still be taking place even though you tried to 
avoid it.  The ARC size might be too small (by default) under FreeBSD 
to offer similar caching.

Bob
-- 
Bob Friesenhahn
bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/


More information about the freebsd-fs mailing list