Date: Wed, 09 Feb 2005 10:51:27 -0800 From: "Mark A. Garcia" <mag@hamletinc.com> To: Olivier Nicole <on@cs.ait.ac.th> Cc: freebsd-questions@freebsd.org Subject: Re: vinum in 4.x poor performer? Message-ID: <420A5BAF.4050305@hamletinc.com> In-Reply-To: <200502090352.j193qFEq098451@banyan.cs.ait.ac.th> References: <20050208231208.B94338@ganymede.hub.org> <20050208234602.M94338@ganymede.hub.org> <200502090352.j193qFEq098451@banyan.cs.ait.ac.th>
next in thread | previous in thread | raw e-mail | index | archive | help
Olivier Nicole wrote: >>All servers run RAID5 .. only one other is using vinum, the other 3 are >>using hardware RAID controllers ... >> >> > > >Come on, of course a software solution will be slower than an hardware >solution. What would you expect? :)) > >(Given it is same disk type/speed/controler...) > > > Usually this is the case, but it's also very dependent on the hardware raid controller. There are situations where a software raid (vinum in this case) can outperform some hardward controlers under specific circumstances, i.e. sequential reads w/very large stripe size. An example is an image server where the average image might be 3MB. A stripe size of 434kB would cause ~7 transfers of data. A case for a larger stripe size of 5MB would greatly improve performance. There would be an 2MB diff in the avg file size that doesn't have any useable data. Only 1 transfer of data would occur. Vinum optimizes the data transfered to the exact 3MB of the file, whereas some hardware controls would transfer the whole 5MB stripe, adding some bandwidth latency and transfer time. Again, it's a matter of specific cases, and assuming 'performance' based on differing conduits for data transfer can just skirt the real issue, if there is any. Cheers, -.mag
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?420A5BAF.4050305>