Panasas decides to redefine RAID
"With Horizontal Parity, for example, the company uses multiple RAID controllers to perform recovery tasks in parallel."
. . . so one of their customers had triple disk failures in one raid group - nasty. Does the Panasas O/S do regular "disk scrubs" like one vendor I know does? Was there no predictive failure and copy-out of the "failing" disks onto spares? No, even nastier.
"But basically there are a bunch of codes that we implement on there to detect errors at a sector level."
. . . Would that be something like the scsi "g" list of bad sectors? And anyway all disk writes are checksummed - that's a given.
"RAID 6 or double parity RAID"
. . . Since RAID 6 is defined as RAID 5 with another dedicated parity disk, the data used for reconstruction is literally, all over the place. With NetApp's Double Parity RAID (RAID DP), there are 2 dedicated parity disks per raid group so there's far less overhead on both everyday writes and reconstructions.
Then there's the fact that e.g. NetApp's O/S, Data ONTAP is RAID aware, I don't know of another vendor who can say the same.
"Like the RAID 6 approach, Vertical Parity does require extra overheard in the way of disk space. Panasas contends that its overhead - 20 per cent - equals that of RAID 6."
. . . NetApp's default raid group size for RAID DP is 16 disks, 2/16 = 12.5% overhead.
"But Panasas does not require more space as the disks grow in density, while the RAID 6 crowd does."
. . . and there was me thinking that 20% of a bigger thing was bigger than 20% of a smaller thing.
"With Network Parity, Panasas performs a complete check on data as it moves between storage boxes and server/client systems."
. . . That will be something like the OSI model's layer 3 then - TCP
Rosenthal seems to be talking out of his bottom.
Owen