Re: Orion: UFS vs ZFS testing questions

From: Martin Klier <usn_at_usn-it.de>
Date: Thu, 06 Mar 2008 13:19:35 +0100
Message-ID: <47CFE157.7040909@usn-it.de>


Hi Michael,

Michael Schmitt schrieb:
> I have some questions about testing I did using the Orion tool against
> UFS and ZFS. We found that UFS was giving us better IOps results for
> small IO, while ZFS was giving us better MBps for large sequential IO.
> I was hoping that the list can help me figure out why. First, are there
> any issues using Orion to test these file systems?

maybe ZFS checksums and the internal transaction procedure / copy-on-write of ZFS kills your IOps performance. If you are able to read German, you might want to look up this papers.

http://www.linux-mitterteich.de/uploads/media/ZFS_DatenSich.pdf http://www.linux-mitterteich.de/uploads/media/ZFS_DatenSich_presentation.pdf

Summary:
Block checksums for each block are calculated and kept through all locations the block resides in, and checked at read time. The transaction system with copy-on-write causes multiple IOs per IO (it's a kind of chain reaction when higher organizational units (like the uberblock) are changed).

Regards
Martin

-- 
Usn's IT Blog for Linux, Oracle, Asterisk
www.usn-it.de

--
http://www.freelists.org/webpage/oracle-l
Received on Thu Mar 06 2008 - 06:19:35 CST

Original text of this message