Date: Thu, 19 Jul 2007 20:47:34 +0100 From: Doug Rabson <dfr@rabson.org> To: "Mark Powell" <M.S.Powell@salford.ac.uk> Cc: freebsd-fs@freebsd.org Subject: Re: UNS: Re: ZfS & GEOM with many odd drive sizes Message-ID: <200707192047.34979.dfr@rabson.org> In-Reply-To: <20070719203134.B4923@rust.salford.ac.uk> References: <20070719102302.R1534@rust.salford.ac.uk> <200707192027.44025.dfr@rabson.org> <20070719203134.B4923@rust.salford.ac.uk> (sfid-20070719_20342_28B5762E)
next in thread | previous in thread | raw e-mail | index | archive | help
On Thursday 19 July 2007, Mark Powell wrote: > On Thu, 19 Jul 2007, Doug Rabson wrote: > > On Thursday 19 July 2007, Mark Powell wrote: > >> Should I expect much greater CPU usage with ZFS? > >> I previously had a geom raid5 array which barely broke a sweat > >> on benchmarks i.e simple large dd read and writes. With ZFS on the > >> same hardware I notice 50-60% system CPU usage is usual during > >> such tests. Before the network was a bottleneck, but now it's the > >> zfs array. I expected it would have to do a bit more 'thinking', > >> but is such a dramatic increase normal? > >> > >> Many thanks again. > > > > ZFS does a checksum on every block it reads from the disk which may > > be your problem. In normal usage, this isn't a big deal due because > > many reads get data from the cache. > > I've turned off checksums, but still my machine is struggling. I > think my Athlon XP is a little old for all this work :( Any other > tips for speeding zfs up? > Cheers. Nothing really comes to mind. You could try simpler geometries (e.g. mirrors or collections of mirrors). Having at least some of your drives in a simple configuration might be useful - I'm working on ZFS boot code at the moment and I don't intend to support raidz or raidz2 (at least to start with). Collections of mirrors and simple disks are much easier.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200707192047.34979.dfr>