In-Depth

NetApp and Grid Storage: More Than Just Marketing?

Is Network Appliance's grid storage campaign just marketing fluff?

• By Jon William Toigo
• 09/02/2004

Grid computing is a technology for dynamically allocating and de-allocating capacity, bandwidth, and processing based on the demands of an application. It is what numerous colleges and universities (and more than a few national research labs) are doing to obtain supercomputing capabilities on a Linux white box budget.

So, what, then, is grid storage? NetApp’s marketing blitz says that it is a way to scale out, rather than scale up, NAS storage using relatively inexpensive storage building blocks. They propose to salvage the best from their existing Filer technology (Berkeley Fast File System-based Write Anywhere File Layout or WAFL, plus considerable brand recognition) and mix it together with secret sauce that they obtained with their purchase of Spinnaker Networks (Beowulf-clustered NAS heads, plus an Andrews File System-based extensible file system), to enable NAS to scale horizontally and without the management pains and file system addressing limitations usually associated with such an architecture.

According to Bennet, the company will use Spinnaker to create a virtual head from many individual NAS heads, using a 200-series machine as something approximating a RAID controller. Secondly, the company will “parachute WAFL and other data management tools into the Spinnaker technology.” He concedes that a multi-year process will be required to realize a “converged codeline” between the products.

NetApp’s marketing also defines grid storage as a strategy that embraces an implementation of “tiered back-end storage” intended to leverage low-cost near-line and offline storage repositories to host data that is less frequently accessed and/or modified, and they are planning to eventually tie this scheme back to applications themselves so that storage can be purpose-built to meet the needs of specific apps. Result: Network Appliance gets into the Information Lifecycle Management business.

Bennet says that this architecture will solve “overarching storage problems” and will move administrators out of the disk management business and back into the application management business. “Current network storage technology,” he says, “is like surfing the Net, but needing to know the MAC address of every connected node.” NAS provides a simpler management model than contemporary SAN, and using NAS as a gateway to a SAN will provide the means to better manage the back-end storage complexity.

While we quite agree with Bennet’s characterization of the messy world of SAN management, and long ago opined in this piece that a scalable cluster of NAS heads would provide better NAS scalability and improved back-end storage management, we see a deeper meaning to grid storage that only flickers in the background of the current NetApp vision: That is the creation of a limitless file system address space with support for heterogeneous file sharing.

Given the enormous incompatibilities between NetApp’s Berkeley Fast File System implementation and the Andrews File System embedded in Spinnaker technology, one immediate question is whether NetApp can deliver a converged-codeline product that is downwardly compatible with the company’s existing product line. The second is whether the engineers are ready to tackle the systemic problems of file sharing and synchronicity that we all confront.

I'll have more to say on the subject in my next column. For now, please send your comments to jtoigo@intnet.net