On Thursday November 13, bugzilla@watkins-home.com wrote: > I am setting up an array with 14 disks. > Should I create one 14 disk RAID 5 array? > Or two 7 disk RAID 5 arrays, and then RAID 0 them together? > I know I would have less overall space with 2 RAID 5 arrays. This is not > an issue. > > I guess the real question is: does RAID five have a sweet spot related to > the number of disks? > > Is there chunk size sweet spot? Does it very with number of disks? > > System: P3-500 (2 processors) > 512 Meg ram > 3 SCSI cards for disks. > 1 internal LVD (80 meg/second) > 2 disks (for OS) mirrored > 1 disk for spare > 1 external LVD (80 meg per second) > 7 disks > 1 external ultra-wide (40 meg per second) > 7 disks Sweet spots are very system dependant. Given the different performance characteristics of the two busses I would make a raid5 out of each set of 7 disks and then combine the two sets with raid0 or linear. > > I could not find any performance info related to this subject. > Also, could not find much about chunk size. > > I did a simple dd test of these disks using a block size > of 16, 32, 64, 128, 256, 512 and 1024K. 16 and 32K were best for overall > speed and CPU usage. It got worse as the block size increased. I work with > HP-UX systems alot. On HP systems, as block size increases CPU load decreases. > But HP-UX has raw (character) devices. My dd test was with RedHat > 8.0. A dd test on the disk with a given blocksize is very different than a dd test on a filesystem on a raid array on the disks. When doing tests it is *always* best to make the configuration and load as close as possible to what you really plan to do as there are many variable that can distort the outcome. I find a chunksize of around 64k - 128k works pretty well. Bigger chunk sizes don't seem to give significant improvements. NeilBrown - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
