It won't be your father's filer. Network Appliance Inc. (NetApp), granddaddy of NAS, has embarked on an ambitious project that fundamentally rethinks its storage architecture. Over the next couple of years, Sunnyvale, Calif.-based NetApp will introduce storage that departs radically from its traditional file-centric monolithic controller-array pairs. "Storage Grids," as NetApp is calling them, are clusters of highly scalable storage controllers that are physically de-aggregated from their back-end disk resources.
The clustering technology that makes this all possible comes by way of Spinnaker Networks, which
NetApp's Storage Grids will parallel an important trend in the server world: the move away from large symmetric multi-processing (SMP) systems, and toward large server farms made up of commodity Linux servers. That trend is quite evident today in the technical computing space, where NetApp intends to find its first customers for the Storage Grids technology.
When placed in those environments, today's storage technology "begins to look archaic," says Vasudevan.
Clustering, though, is only one part of the Storage Grids equation. Other "building blocks" NetApp is working on have been in the hopper for several years already, Vasudevan says.
For example, NetApp believes its customers want to be able to optimize existing storage resources. That can be achieved with several technologies. On the one hand, NetApp has already delivered on its unified storage architecture, offering the ability to run either file or block applications on its systems. Then, sometime in 2005, NetApp plans to introduce compression and data de-duplication functionality that works by recognizing patterns in block-level data. Theoretically, that should help IT administrators reduce the overall amount of data that needs to be stored.
Another key component of a Storage Grid is the ability to perform what NetApp terms "non-disruptive data grooming." In other words, data in a Storage Grid should be able to move from one node to another without affecting applications. This ability is a cornerstone of any information life cycle management (ILM) product, Vasudevan says.
To that end, NetApp engineers are working to expand the company's vision of a global namespace. The global namespace initiative, Vasudevan says, "is where most of our investment is headed."
As it stands, NetApp already has a "file-centric global namespace," Vasudevan says, but it "will look very different when it is applied to blocks."
Two areas that NetApp won't need to touch are its Write Anywhere File Layout (WAFL) and its highly esteemed RAID-4 implementation.
Storage Grids are predicated on a technology bet: that networks, specifically IP networks, will continue to get faster and cheaper. In a clustered environment, the bandwidth required to satisfy inter-node communication is substantial. Existing nodes in Spinnaker clusters communicate over Gigabit Ethernet, but in the future, that connection may happen over 10 Gigabit Ethernet running RDMA (remote data memory architecture), Vasudevan says. At 10 Gig speeds, Ethernet may be fast enough "to replace the internal bus."
Analyst Randy Kerns of the Evaluator Group says that Storage Grids are "good stuff," but he still has his doubts. "Today, the only markets where [Spinnaker-style scalability and performance] are wanted, needed or requested are national labs," Kerns says. And "it's arguable how big that market is and how long it will take for it to become mainstream."
But Kerns concedes that if and when massive scalability and performance do become mainstream needs, "they'll have a product for them."
FOR MORE INFORMATION: