Raid 0+1 Versus Raid 5?

[bluesherpa]

Ran into this article: http://blogs.zdnet.com/storage/?p=483

How accurate do you guys think that is? Do most Raid 5 controllers stop building the array on a single drive read error?

 

[icrf]

I think it's true. There's an older article from 2007 from ZDNet saying 2009 is the year RAID 5 dies, which I think is the one that started all this.

I use linux and mdadm in an older computer case to do my raid 5, and it has more options. I had a controller card flake out on me and three drives (out of nine) dropped out of the array at once. The three drives were marked bad and I thought all was lost. As it turns out, there is an "--assume-clean" option that let me reassemble the array with no issues, despite the drives technically being out of sync.

But, more to the point, so far as I know mdadm will still fail when it reaches a read error, though I have hope that it can be edited to ignore such things. I really should move up to raid 6. I'd rather run zfs, but it doesn't support online capacity expansion. Btrfs will probably be the one to have, but it's pretty new and is still only alpha quality.

 

[eon]

Hardware raid5 it is true.

For Hardware raid 5 it is true. Many software raids have workarounds. ZFS for example will still allow only one disk to fail on raidz=raid5 (but with no write hole) but it allows you to allocate a hot dynamic spare.