On Thu, 24 Jul 2014 10:43:01 Toby Corkindale wrote:
On 24 July 2014 09:45, hannah commodore <hannah@tinfoilhat.net> wrote:
I've inherited a system which has an Adaptec 2420SA RAID controller and 4x SSDs running in RAID6. There's no end to problems experienced with this setup; a disk seems to time out and get ejected from the controller at least every week. It's not always the same disk though, and I've tried replacing each of the SSDs with their spares, to no avail.
This could be related to TRIM. If you have it enabled it can cause significant delays and if you don't then eventually the SSD will need to clear erase blocks and cause delays for random writes.
Why the hell would someone choose to create a four-disk RAID6 array? It doesn't make any sense!
http://en.wikipedia.org/wiki/Linux_software_RAID Sure it does. A RAID-10 can be made from 2*RAID-1 arrays in which case there's a 1/3 probability that a second disk failure will lose data. If a RAID-10 has every data block on 2 disks with the duplicated data striped (as described in the above Wikipedia page) then if you lose 2 disks then you are guaranteed to lose data. Now if your RAID supports a "replace" operation (where constructing the new disk takes data from the old disk OR the parity disks) then you could lose 2 disks and probably not lose data (most disk "failures" don't involve total loss of disk function, thousands of bad sectors in a terabyte disk is a significant failure). That's good for people who use BTRFS or ZFS. People who use Linux software RAID and other less capable RAID systems need to remove one disk from the array before they add a replacement, so any error found during the process of regenerating the replacement disk will lose data. -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/