Date: Fri, 25 Dec 2009 13:54:54 -0600 From: "James R. Van Artsdalen" <james-freebsd-fs2@jrv.org> To: Solon Lutz <solon@pyro.de> Cc: freebsd-fs@freebsd.org Subject: Re: ZFS RaidZ2 with 24 drives? Message-ID: <4B35188E.7010602@jrv.org> In-Reply-To: <1696529130.20091223212612@pyro.de> References: <568624531.20091215163420@pyro.de> <42952D86-6B4D-49A3-8E4F-7A1A53A954C2@spry.com> <957649379.20091216005253@pyro.de> <26F8D203-A923-47D3-9935-BE4BC6DA09B7@corp.spry.com> <1696529130.20091223212612@pyro.de>
next in thread | previous in thread | raw e-mail | index | archive | help
Solon Lutz wrote: > I opted for two 12-disc raidz2. > Reasons were: Space is more important than performance. > > But performance is very poor - have a look at the iostats Try a RAIDZ with fewer drives - maybe four drives - and see if performance is better. 12 drives in RAIDZ2 may result in a stripe that is just too big for the writes your tests do. > sometimes nothing really seems > to happen for up to ten seconds, or very little data gets written. Might this be a problem > of the amd64 system having only 4GB of RAM? Any tuneable sysctls? Enabling prefetch didn't help...: Any copy-on-write filesystem is going to greatly benefit from very aggressive write-deferral and combining. The cache probably isn't flushed unless it needs to be, or until the next transaction group commit. "10 seconds" is roughly the usual interval for ZFS commits. This is likely not a problem.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4B35188E.7010602>