On Tue, Dec 09, 2003 at 02:22:52AM +0100, Xose Vazquez Perez wrote: > Hans Müller wrote: > > >>but raid 5 only supports a failure in *one* disk, and the rebuild of the > >>raid is _very_ slow. There is big degradation if one disk is broken. > >>Raid 10 can supports 2 failures in differents set of raid_1 and degradation > >>with 1 disk bad is less than raid5. > > > tht is not correct. when one disk in each arry fail the all data are lost. > > only 2 disk or more in the same arry can fail > > I don't know what you mean. > > RAID_5 - _only_ one disk can fail. > RAID_10 - it's two raid 1 joined by a stripe and can fail one disk in every > raid 1 set. He's describing, what I've heard referred to as RAID 0+1, you are describing what I have heard referred to as RAID 10. The difference is if the mirrors are above or stripes. If you take 8 disks, create four mirrors, and 1 stripe of the four mirrors that's RAID 10, and you can have two disks fail as long as it isn't both halfs of the same mirror. If you take 8 disks and create 2 stripes and 1 mirror from the two stripes, that's RAID 0+1, if you have one disk fail, half of the disks are dead (that RAID 0 fails, which causes one of the mirrors to fail). If you have a choice, RAID10 is almost always better in terms of redundancy, but it's my understanding that at points, RAID 0+1 can perform better depending the precise nature of your setup. (In my case, I can't do RAID 1 across controllers, so I do RAID 0+1, where the RAID0 is in hardware, and the RAID1 is between the two controllers). RAID 5 isn't so bad if you do lots of reading or big streaming writes, and failures are relatively uncommon. They suck if you do lots of small writes all over the disks. In theory, it's possible to build a RAID 5 that can withstand multiple failures (using Hamming codes), but I don't believe Linux supports that. > > > -- > fedora-list mailing list > fedora-list@xxxxxxxxxx > To unsubscribe: http://www.redhat.com/mailman/listinfo/fedora-list >
