Date: Thu, 04 Mar 2004 10:21:59 -0500 From: Chuck Swiger <cswiger@mac.com> Cc: freebsd-questions@freebsd.org Subject: Re: 1 processor vs. 2 Message-ID: <40474997.60403@mac.com> In-Reply-To: <200403040050.53556.danny@ricin.com> References: <BC6A0533.1DC4C%joe@jwebmedia.com> <20040303222612.W39053@guldivar.globalwire.se> <20040303213641.GA37555@happy-idiot-talk.infracaninophile.co.uk> <200403040050.53556.danny@ricin.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Danny Pansters wrote: [ ... ] > Physical disks are your unit of failure or of resilliance if you like. Absolutely--- understanding RAID properly requires understanding the division of data onto the physical disks. This is an important concept. > That's why you need 5+ drives for RAID5 to be any real fun. A three-disk RAID-5 array is a devolved case, but it will function usefully. I don't disagree: RAID-5 is best suited for 5 to 7 disks, but is okay with 4. > You want your data to be present at least twice on different physical > drives. You want the same for your parity info. You're actually describing RAID-1 or RAID-51 here. > The mere fact that you stripe everything out with RAID5 > doesn't change your physical unit which is one disk. Resilliance means: what > happens if a random drive fails. RAID5 on 3 disks defeats the purpose of > RAID5 IMHO. RAID-5 means "stripe set with rotating parity", and it is the usage of parity information that provides redundancy (what you call resilliance). A 3-disk RAID-5 array will still function without data loss even if any of the physical disks fails. RAID-5 will not survive a two-disk failure, whereas some RAID-1 or RAID-10 arrays can survive loss of multiple disks. (As might RAID-51...) > Theoretically the more drives, the better RAID5 gets, so that > might say something about Veritas if they warned against using more than 7 > drives. Perhaps grog can be the final referee here, not my turf ;-) The parity calculations for RAID-5 are a lot of work and that work scales linearly with the number of drives in the array. The longer you make the array, the worse the performance becomes for small writes in particular. Very long RAID-arrays also tend to encounter bottlenecks with the system bus or the transport bus rather than being bottlenecked by the drives themselves. > I just kinda fell back into the developed thread, hope you don't mind me > adding a general remark: One doesn't do RAID to increase performance. Period. Sorry, but wrong. The only purpose of pure RAID-0 is to increase performance by distributing I/O between more than one device. Modes using RAID-1 mirroring will also show a significant improvement to read performance. There are three factors involved in deciding how to utilize RAID: reliability, performance, and cost. One can choose to prioritize one or two of these factors, at the expense of the third; which factors are chosen determine which RAID configuration is appropriate. RAID-5 does not increase performance because the priorities are improving reliability (ie, gaining fault-tolerance) and minimizing cost (ie, getting [n-1] available storage from n drives, rather than [n/2] as per mirroring). -- -Chuck
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?40474997.60403>