Taking RAID out of the box

A reader that identifies himself as "Layman" suggests that all the fuss I made in my blog about the reliability of disk drives is somewhat old school.

[ Mario  Apicella's column is now a blog! Get the latest storage news from the Storage Adviser blog. ]

Layman raises a good point, because it's true that clustered storage solutions can fly with aplomb over dangerous situations (such as multiple drives kicking the bucket at the same time) and stay afloat even if a node becomes another entry in the early-mortality statistics of hardware components, as happened here.

Nevertheless, I have to ultimately disagree with Layman. Disk drive reliability is still important, no matter how dependable storage systems are. Even when it doesn’t cause data loss, a drive failure is more than a minor annoyance — at a minimum, it can cause slowdowns and divert precious storage admin time from more productive tasks.

Besides, clustered storage remains a minuscule percentage of all the storage systems sold, although I'll admit that fact may be changing as more clustered systems (usually from small vendors, but that’s a topic for another day) challenge the status quo of monolithic storage.

One of the latest interesting additions to the very diverse landscape of clustered storage comes from a startup with an intriguing name: Pivot3. Pivot3 just came out with the RAIGE (Redundant Array of Independent Gigabit Ethernet) clustered storage system.

It may have an aggressive-sounding name, but the goal should be clear: In essence, RAIGE brings the reliability of RAID to a higher level that covers not just disk failures inside a single node, but across nodes.

Pivot3 calls its 2U, rack-mountable nodes — filled with up to 12 times 500GB or 750GB SATA drives — Databanks, each of which runs a proprietary, Linux-inspired OS that works in cooperation with its siblings on other nodes to make RAIGE a system that largely administers itself. For example, adding a new node automatically expands capacity to include the additional drives and takes advantage of the new GbE (Gigabit Ethernet) connection to improve transfer rate.

Like other clustered systems, RAIGE doesn’t require admins to worry about the physical location of data. Instead, the system keeps track of where each volume’s blocks reside and routes I/O operations accordingly. RAIGE lets admins define RAID level 0, 1, or 5, but with the additional twist that when a failure happens, all drives participate in rebuilding the volume. That translates into much faster recovery time than with traditional RAID.

I haven’t had a chance to touch a RAIGE system yet, but I like that it supports the faster performance and better reliability of multipath I/O across its iSCSI, Windows, and Linux clients. Also worthy of appreciation: Each Databank has dual power supply (good!) and battery-protected cache memory.

In addition — and this is a rather unique feature — Pivot3 reserves strategically located sectors of each disk drive to serve as an additional cache layer. Those dispersed cache sectors should speed up write performance significantly, because new blocks can be dropped to those sectors almost immediately and without having to wait for a complete rotation of the platter.

At the moment, Pivot3 is targeting the video surveillance market with RAIGE, a segment where its attractive price/performance ratio should make a killing, I am told. By year's end, it will open to other markets, which I'm guessing gives the company time to complement its very promising platforms with more management applications such as snapshots, for one.

Join me on The Storage Network with questions or comments.