Microsoft's Boxwood research project could restrict the SAN market according to a report in EE Times. The project aims to replace notions of physical, virtual or logical disks as pointers to storage with what it calls higher-level data abstractions instead.
The Microsoft project has found that such an approach reduces the storage bookkeeping carried out by client systems. It has also revealed that structural information is inherent in the data abstraction and enables better load-balancing, data pre-fetching and caching. The storage system would implement these items once rather than have them replicated in each client system.
Microsoft talks about a linked B-Link Tree data abstraction, a variation of a B-Tree. In a B-Tree the number of decisions needed to locate a desired data item is reduced compared to a binary tree, which has two choices at each level in the tree. A B-tree can have many choices. It means more work has to be done at each decision point but fewer decisions are needed overall, particularly as the number of data items grows.
Microsoft conceives of a collection of server nodes implementing the B-Link Tree, ranging from disk units, through disk controllers to ordinary PCs. There could be a networked set of PCs with disks attached by SCSI or IDE. A four-node cluster of PC machines with GigE links and hot-pluggable disks has been built to demonstrate and pilot Boxwood. A multi-node NFS server is being built that uses the Boxwood cluster as its basic store instead of a local file system.
The EE Times report suggests that users could cobble cheap systems together to function as commercial data centres. Storage tasks could be handled by clusters of thousands of servers each with their own disk drives.
To this writer the idea seems to be utter bilge.
The idea that clusters of thousands of PCs could function as a storage resource is either startlingly trivial or seriously silly. If the PCs are handling other work as well, functioning as application servers of some kind, then having them function as a storage infrastructure using their local disks pooled into some univeral virtualised store would be a performance and management nightmare.
If they are dedicated just to the storage work then all we have are clever storage controllers linked together. We already have that in high-end storage products from companies such as 3Par and Exanet. Having such systems use a B-Tree type of abstraction method rather than the traditional file systems to represent the storage to client systems is, well, ridiculous.
Microsoft is one company that has benefitted enormously from legacy computing ideas. The massive installation of PCs using Windows has shown the power of the installed base as Windows has progressed through its iterations. The idea that the world's population of server users will rewrite millions of applications changing them from filesystems to B-Link Trees or some other abstraction is simply ludicrous; seriously silly.
If the B-Link Tree abstraction is implemented transparently, inside disk controllers clustered together for example, then the impact of this research will be largely hidden. The research is being carried out by a small team of post-graduate IT people and hasn't even been written up in a first formal paper yet. Do not hold your breath. Your SANs and NAS are safe.