Author
|
Thread
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
LightningCrash
Smile like Bob, order your free LC today
Joined: 03 Apr 2003
Posts: 5020
|
|
|
Oh I have to revise my statement.
The behavior of the RAID 5 was historically as follows:
If 1 Drive is lost and URE occurs during rebuild, stop rebuilding.
The behavior of what some vendors were touting as RAID6 during that time was as follows:
If 1 Drive is lost and one URE occurs, offline the first parity stripe and rebuild with the other parity stripe. If another URE occurs, stop rebuilding.
Today:
Most RAID5 firmware will just offline the data stripe in particular if a URE occurs during rebuild. The rest of the array will rebuild.
RAID6 today attempts to rebuild with the XOR stripe. If the XOR stripe does not compute for some reason, the ECC from the second non-data stripe is tried. The second non-data stripe isn't parity bits (It is unlike the first, XOR stripe), it's generally Reed-Solomon codes. This is similar in many ways to what PAR2 does (it also uses Reed-Solomon) for Usenet/BitTorrent.
So the odds of having a URE somewhere on the data or XOR stripe are pretty low already... but the odds of having a URE on the adjacent Reed-Solomon stripe are even lower, by orders of magnitude.
Desktop drives generally have a URE rate of 10^14. There are some exceptions here, namely the WD20EADS and similar drives. They have a URE rate of 10^15.
I would build RAID6 if it were general storage.
The important thing to know is that this data is not forever.... this is a transitive state.
Storage is not archival and vice versa.
In many ways, archival of important information is more important than online storage of important information.
|
Fri Jan 07, 2011 3:50 pm
|
|
|
|