[NCSA-discuss] RAID recommendations?

Dewey Hylton dewey at hyltown.com
Fri Dec 17 23:22:53 EST 2010


On Dec 17, 2010, at 10:37 PM, Kevin Wilcox wrote:

> On 17 December 2010 17:59, Steve Wolfe <swolfe100 at gmail.com> wrote:
> 
>> Simon's suggestion looks interesting - and then maybe run sotware raid
>> and maybe LVM.
> 
> <snip>
> 
>> A few years ago we ran some software raid
>> and although not truly ideal .. not too bad.
> 
> Given what we've heard about the scope of the project, software RAID
> sounds like a great solution.
> 
> I'm really surprised that, on a SysAdmin list, nobody has suggested a
> machine running either
> 
> a) FreeBSD + ZFS or
> 
> b) Solaris + ZFS
> 
> given that ZFS handles both the filesystem component and the RAID
> component with error checking, snapshots, etc., and is, generally
> speaking, a lot more "user friendly" in a bad situation (failing or
> degraded disks) than LVM.
> 
> kmw

i concur, and recommend solaris + zfs. i've seen and been in some ugly failure situations that zfs has handled beautifully. 

my favorite mass storage configuration with zfs is basically raid-10, striping across mirrors. with zfs, this seems to give the best of all features - at least for my uses. here is a real-world example:

8 hot-swap sata drives, 300GB each. 4 mirrors, striped. this yields about 1TB of storage considering overhead and the 1024 vs 1000 argument. this array can theoretically withstand 4 simultaneous disk failures. i've lost 3 within a week, so this makes me happy. anyway, yank a disk and replace it with a 1TB disk. or better yet, yank and replace 4 (one from each mirror). wait for the zpool to resilver (this is faster than you might expect). then yank and replace the other 4. wait for the resilver. export/import the zpool and you have 4TB of usable storage.  or just replace them with larger disks as they fail, and let it grow slowly. you can add another couple of mirrors and grow wide. and you can do all this on-the-fly, while the storage is in use. yup, i'm in love.

and that's just zfs - combined with nfs or comstar or whatever and you really have something.

zfs seems fairly stable on freebsd as well, and obviously cheaper than solaris - but i trust the solaris zfs more so that's what i use. plus, solaris zones rock.



More information about the ncsa-discussion mailing list