Andrew Piskorski wrote:
I believe Garth's whole point is that your assumption above is often
NOT true. He also seemed to imply that this is a function of the
ineraction between the block-level RAID implementation and the file
system, as his Panasas file system reputedly fixes this scary, "one
small unrecoverable read during array rebuild kills your entire disk
volume" failure mode.
Panasas is an object-based filesystem. Individual files are stored as
either a mirror (small files) or as a RAID-5 across the filesystem.
If you should be in the middle of a reconstruction process and THEN run
into bad blocks only the file which contains those bad blocks is affected.
Panasas also has "distributed sparing" which is kind of interesting.
Instead of allocating a complete spare disk (or disks) in a RAID array,
the filesystem allocates a certain percentage of space as spare space -
a high water mark if you will. So there is no need to allocate
individual spare blades, and the user can choose how much spare capacity
is desirable (say you have ten storage blades, then 10% of space would
be allocated. If you have 20 blades, and specify one blade's worth of
redundancy you have 5% allocated)
When a reconstruction process starts this spare space is used up.
When a failed blade is replaced with a new one, the filesystem migrates
to it over time, until the N% sparing is restored.
_______________________________________________
Beowulf mailing list, Beowulf@beowulf.org
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
