By Dave Simpson
—Panasas, which focuses primarily on the high-performance computing (HPC) market, this week introduced a twist on standard RAID implementations that addresses the problem of disk media errors, as opposed to only disk-drive failures, in high-capacity RAID systems—a problem that has been exacerbated by the advent of very-high-capacity (e.g., 1TB) disk drives.
First, a little background, adapted from Panasas collateral:
Traditional RAID implementations protect against disk-drive failures by calculating and storing parity data. Over the past decade, disk drives have become about 10 times more reliable, yet more than 250 times denser. Unfortunately, the number of disk media failures expected during each read over the surface of a disk grows proportionately with the increase in density, and has now become the most common failure mode in RAID.
A RAID disk failure can cause loss of all the data in a volume, which may be tens of TBs or more. Recovery of the lost data from tape can take days or even weeks. Some RAID vendors address this problem with RAID 6, or double-parity RAID, which protects against the simultaneous failure of two disk drives. However, these implementations only address the problem at the drive level, and do not address the issue of media errors.
Panasas' Tiered-Parity architecture consists of three levels of error detection and correction:
- Horizontal parity, which is based on the company's existing ObjectRAID RAID-5 technology;
- Vertical parity: New in the 3.2 release of Panasas' ActiveScale operating environment, vertical parity is essentially RAID implemented at the disk-drive sector level, as opposed to the disk-array level (hence Panasas' term "RAID-on-Disk"). Vertical parity detects and fixes media errors on disks before the errors can interfere with RAID recovery (rebuild). Although vertical parity requires about a 20% capacity overhead (the same as RAID 6), Panasas claims there is no performance penalty associated with vertical parity.
- Network parity rounds out the end-to-end data integrity goal by guaranteeing the integrity of data that goes across the network to Panasas' ActiveStor Parallel Storage Clusters.
(Note:One of the original "inventors" of RAID—Garth Gibson—is CTO at Panasas. RAID was first implemented 20 years ago.)
For more information on RAID see "Trends in RAID drives, interfaces, controllers"