Panasas will include something called vertical parity in version 3.2 of its ActiveScale clustered file system (CFS) software. The feature uses open standards-based Error Correcting Codes (ECC) to calculate parity algorithms for the data on each sector in a disk drive so data can be reconstructed if one sector becomes corrupted. Version 3.2 is slated for beta testing by November and expected to ship by year-end.
It's a problem not addressed by dual-parity RAID 6 systems, which use two separate disk drives to protect against double whole-disk failures, according to Gibson. "Rebuilding in a RAID 6 scheme is meant to prevent against double whole-disk failures, not a media read error in one sector of the drive," he said.
With 1 TB drives just beginning to ship, vertical parity might be overkill for all but the largest of storage systems, according to Data Mobility Group senior analyst Robin Harris. However, "today's HPC environments are tomorrow's enterprise," he said.
Harris cited the rise of social networking and collaboration in the business world through applications like SharePoint and private Web portals, as well as the impending approval of the parallel NFS (pNFS) standard by the T11 standards committee, as harbingers of a new trend in the enterprise. "We're going to see a lot more HPC-style I/O in the enterprise in the next five years," he said. "And larger and larger file sizes, within larger and larger amounts of unstructured data."
Eventually, Harris said, RAID 5 algorithms will cease to be effective. So who can design the most effective next generation of dual parity to replace it? So far Panasas has as good a shot as anyone, especially since vertical parity, if it works, only hogs 10% of the surface of the disk, rather than requiring the addition of two additional drives' worth of capacity.
"The problem with RAID 6 is that it can be very expensive insurance, requiring two drives' worth of power, cooling, floor space, rack space and capacity," Harris said. But vertical parity remains an unknown in real-world environments until users get the chance to put it through its paces.
Developers focus on RAID
Other vendors are also tackling the limitations of RAID. For example, Network Appliance Inc. (NetApp) has changed the way storage controllers in its boxes read data on disk to reduce RAID 6 I/O loads. Overland Storage has jumped into the RAID 6 space with a feature called DriveAlive, a kind of hot-swapping feature that staves off full-rebuild mode when one of the disk blades in the array chassis is removed for routine maintenance. NEC Corp. of America began shipping a grid-based product, HydraStor, six months ago. HydraStor's Distributed Resilient Data (DRD) has a default setting of Parity 3 protection. That guards against three disk failures, but users can effectively "dial up" as many parity chunks as they require.
Several startups have also recently emerged from stealth with new ideas about grid architectures for data protection, bypassing the idea of RAID altogether. Pivot3 released its first storage product based on its RAID Across Independent Gigabit Ethernet (RAIGE) IP. Parascale Inc. unveiled software designed to turn commodity servers into massively parallel storage clusters.