In <20090324080547.GA6422@saturn.hollstein.homelinux.org>, Manfred Hollstein wrote:
On Tue, 24 Mar 2009, 06:00:35 +0100, Arun Khan wrote:
I am wondering if is there any significant advantage of doing RAID 1+0 (4 disks) v/s RAID5 (3 disks).
RAID 1+0 on 4 disks is going to be faster than RAID 5 on 3 disks. Not a lot mind you, but it will be faster. There are also a few failure modes where 1+0 can recover from 2-disk failure where RAID 5 never does, so it does have a slight reliability advantage.
I never actually used RAID5,
I have used RAID 5 and RAID 6.
but I read everywhere that write performance is quite bad (confirmed by the document above);
Not in my experience.
but this is not surprising, as calculating the checksums for each block comes with a cost, of course.
The RAID 5 "checksum" is a simple XOR. It's is not even as complicated as CRC-16. I've never once seen it be the bottleneck in pushing data to a RAID 5 or 6 array, even when it is implemented in software. If it is implemented as software heavy disk I/O *will* cause CPU use, so it can make the system perform worse when both disk and CPU are under heavy load. RAID 5 on 3 disks is roughly as fast as RAID 0 across 2 disks. RAID 6 across 5 disks is roughly as fast as RAID 0 across 3 disks. You may be writing more data, but the disks can write in parallel, as long as you don't saturate your bus. -- Boyd Stephen Smith Jr. ,= ,-_-. =. bss@iguanasuicide.net ((_/)o o(\_)) ICQ: 514984 YM/AIM: DaTwinkDaddy `-'(. .)`-' http://iguanasuicide.net/ \_/