Date: Fri, 10 Jul 2009 09:11:21 +1000 From: Andrew Snow <andrew@modulus.org> To: "Patrick M. Hausen" <hausen@punkt.de> Cc: freebsd-stable@freebsd.org, "Tonix \(Antonio Nati\)" <tonix@interazioni.it> Subject: Re: ZFS - thanks Message-ID: <4A567919.8000704@modulus.org> In-Reply-To: <20090709134302.GA50485@hugo10.ka.punkt.de> References: <20090709112512.GA44158@hugo10.ka.punkt.de> <73a41d4b72d62b0bfe3d0fb7206376a8.squirrel@cygnus.homeunix.com> <20090709123434.GC46563@hugo10.ka.punkt.de> <4A55EEEC.3030007@interazioni.it> <20090709134302.GA50485@hugo10.ka.punkt.de>
next in thread | previous in thread | raw e-mail | index | archive | help
Patrick M. Hausen wrote: > You cannot escape the poor write performance of RAID 5 and > comparable setups with or without hardware. No matter how > much you cache, one time a block must be written to disk. ZFS RAIDZ works differently: It is based on variable-sized blocks written to the disks based on incoming data stream, grouped into transactions. This makes it very efficient for clustering multi-threaded random I/O writes together into large physical disk writes. (The downside is it has to read the entire "stripe" even if you are only reading one byte, in order to calculate and verify the checksum.) - Andrew
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4A567919.8000704>