Boxwood: an alternative to SANs?

BANGALORE: Microsoft's research project 'Boxwood' has been launched to explore the utility of data structures or abstractions as storage infrastructure.

The recent trends in storage system infrastructure argue for using distribution and virtualization over incremental expansion that brings with itself the problems of scalability, automatic reconfiguration for load and capacity balancing, fault tolerance, and availability.

According to the researchers at Microsoft "data abstractions is an appropriate mechanism to implement storage infrastructure while preserving the benefits of virtualization and distribution.

In contrast to prior work on storage infrastructure that has focused on exposing storage as a simple linear abstraction (either a physical disk, or a logical disk, or a sparse virtual disk), we wish to export higher level abstractions, e.g., trees, hash tables, linked lists. Clients of the system, e.g., file systems, databases, and indexing engines access the system over the network and can create multiple instances of each abstraction. Each instance allows a set of operations that are appropriate for the abstraction."

An online report on Boxwood, says the benefits high-level abstractions could bring with it are of better load-balancing, data prefetching, and informed caching. These mechanisms can be implemented once in the system instead of having to be duplicated in each client. This new approach will allow the developers define the amount of storage they need at a level above block sizes and provide their own identifier for that space, independent of actual storage address space.

"We envision the abstractions to be implemented by a collection of 'server nodes,' where each server node contains a CPU, RAM, disk, and network interfaces. One can think of server nodes ranging from individual disk units, to disk controllers, to ordinary PCs. For example, if we have disks that are directly pluggable into the network, then one can imagine using the CPUs on the disk itself. A more realistic alternative may be to use the CPU in a disk controller that plugs into the network. Yet another alternative is to incorporate the software into a set of networked PCs attached to ordinary disks via SCSI or IDE."

According to the research report there is no one universal abstraction that can serve the needs of all clients. Currently abstractions are being implemented using a B-Link Tree. Each of this tree allows the usual add, delete, lookup and enumeration operations and the same system can then be extended to higher level services like file systems, database engines, and can simplify clustered Internet services. Boxwood uses variations of B-Tree data structures called B-Link trees and can provide global storage- and state-locking mechanisms.

The project if successful, could open the door to wider use of large clusters of inexpensive servers, potentially narrowing the market for storage area networks, say subject experts. According to researchers the project can define the way for future generation of Microsoft's storage virtualization products and will be very attractive to the database developers.

Software developers will be able to handle storage tasks distributed across clusters of thousands of low cost, off-the-shelf servers, each with their own hard disk drives.