Inspeed thinks dig by thinking small - Computer Technology Review Editors Choice 2002

Computer Technology Review, July, 2002 by Mark Ferelli

Routinely, in the SAN switch business, the conventional wisdom is to think big. Switches have been growing to the point that 100-port, director-class switches from the Brocades, McData and InRange have become key products in the implementation of Fibre Channel SANs. But Vixel decided to really innovate, thinking outside the box by thinking well inside the box. Vixel has gone small, creating a 12-port switch on an ASIC to implement within the disk array.

The approach is positively unique, placing the switch inside a RAID array where disk drives are arranged in trays and connected together using FC-AL--which is already a commonplace topology in many RAID storage platforms. Adding InSpeed (basically interconnecting every FC-AL disk tray to a Vixel "chip-switch"), arrays become a switched bunch of disks (Vixel here introduces yet another acronym: SBODs), yielding a three- to five-fold speed improvement. Network Appliance is the company's first OEM, and will be offering InSpeed in upcoming NAS appliances. Many others are evaluating the technology, as of press time.

Vixel's InSpeed Embedded Storage Switches can be applied to improve manageability and performance in two back-end storage switching topologies: As mentioned, they do intra-shelf switching--connecting the drives within the disk drawers, changing the architecture from the standard JBOD (just a bunch of disks) to an SBOD implementation using InSpeed ASICS or Blades. The other topology is inter-shelf switching-connecting the RAID controller to the drive drawers in a backend-switch implementation, using InSpeed Embedded Storage Switches in a box, blade or ASIC format. InSpeed further reduces latency and dramatically reduces multi-hops as read/write requests no longer have to move through each disk drawer before finding the correct drive.

Vixel management advises that this is only the beginning. They hope to eliminate arbitrated loop connections outright, looking for a switch fabric implementation within the array. If they can accomplish this, the speed factor will only be one of the improvements. The ability to bypass a failed drive could be accomplished with much greater simplicity.

One of the most important potentials in InSpeed is that array OEMs will not need to reinvent their controller architectures in order to gain the technology's advantages. In an economic climate where time to market is crucial, this advantage allows potential vendors to get speed and timeliness at the same time. Looks like thinking small yields big results.