On Wednesday 16 June 2010 04:43:06 martin f krafft wrote: > also sprach Boyd Stephen Smith Jr. <bss@iguanasuicide.net> [2010.06.15.2108 +0200]: > > > Use mdadm for a RAID5 or RAID6 and LVM on top for the remaining > > > cases when you need space and care less about performance. > > > > Use RAID 1/0 in mdadm when you need redundancy, space, and performance. > > > > (Although, IME, RAID 5 is not badly performing.) > > Sure it is, on writes. If you have the right hardware, you won't > notice, but every write takes twice as much work. > > http://git.debian.org/?p=pkg-mdadm/mdadm.git;a=blob;f=docs/RAID5_versus_RAI > D10.txt;hb=refs/heads/contrib/docs/raid5-vs-raid10 While disks may have gotten cheaper, some people only need 20% (or less, RAID 5 doesn't have to be 5 disks) redundancy instead of the 50%, 67%, or more redundancy provided by RAID 1/0. (Also, RAID 1/0 in mdadm is not just stripes over mirrored pairs OR mirroring over striped sets.) RAID 1/0 does indeed have all the recovery advantages mentioned, and I generally recommend it when you can afford the 50% redundancy. The support for RAID 3 or 4 over RAID 5 at the end is completely without basis, and almost certainly wrong. RAID 5 performs better than RAID 4 and 3 under all conditions and has the same failure cases. Clearly, the author was biased against RAID 5 from the start, which throws suspicion on the anecdotes in the prose. Still, the issues listed about RAID 5 recovery are very real, and should be enough to make one reconsider either spending more to be able to use RAID 1/0 and have the same usable space or using RAID 1/0 and doing with less usable space. > http://git.debian.org/?p=pkg-mdadm/mdadm.git;a=blob;f=debian/FAQ;hb=HEAD#l > 106 The layout used with near=2 is, of course, exactly the same as layering. It is unclear, then, how this can be applied to an odd number of devices, per the man page: "e.g. it is perfectly legal to have an ’n2’ layout for an array with an odd number of devices" I was under the impression that offset=2 would do better about rotating which pairs were mirrored for a data block. The diagrams show hdd1 being mirrored with hdd2 and hdd4 but not with hdd3; similar with far=2. -- Boyd Stephen Smith Jr. ,= ,-_-. =. bss@iguanasuicide.net ((_/)o o(\_)) ICQ: 514984 YM/AIM: DaTwinkDaddy `-'(. .)`-' http://iguanasuicide.net/ \_/
Attachment:
signature.asc
Description: This is a digitally signed message part.