From mboxrd@z Thu Jan 1 00:00:00 1970 From: Keld Simonsen Subject: Re: messed up changing chunk size Date: Mon, 19 Jul 2010 21:02:46 +0200 Message-ID: <20100719190246.GA11519@rap.rap.dk> References: <4C439D19.4070906@gmail.com> <20100719095351.4536a252@natsu> Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Return-path: Content-Disposition: inline In-Reply-To: <20100719095351.4536a252@natsu> Sender: linux-raid-owner@vger.kernel.org To: Roman Mamedov Cc: Konstantin Svist , linux-raid@vger.kernel.org List-Id: linux-raid.ids On Mon, Jul 19, 2010 at 09:53:51AM +0600, Roman Mamedov wrote: > On Sun, 18 Jul 2010 17:32:25 -0700 > Konstantin Svist wrote: > > > I looked around and found that chunk size > > of 512 should work better. > > Not true, at least with RAID5/6 a chunk size of 64K performs faster, see > http://louwrentius.blogspot.com/2010/05/raid-level-and-chunk-size-benchmarks.html > http://alephnull.com/benchmarks/sata2009/chunksize.html Thanks for these benchmarks, I added them to our benchmarks page on the wiki. It seems odd to me to use sata controllers on a 32-bit PCI bus, this should limit performance to about 130 MB/s - which is also seen tin the graphs. I think people would be better off using the on-board SATA controller, this would give normally something like 4 SATA ports connected to the southbridge, which normally is much faster than a 32-bit PCI bus, and also faster than many PCI-E busses. So I think using all the on-board SATA ports first, and the using the SATA ports on a bus-connceted controller would give better performance. In the louwrentius page, I miss info on which layout of RAID10 that is used. The 3 layouts (near, far, offset) has very different characteristics. best regards keld