SDI Seminar

Speaker: Peter Corbett, IBM T. J. Watson Research Center

The Vesta Parallel File System

Date: March 2, 1995

Abstract: Most multicomputers supply balanced I/O subsystems based on multiple I/O nodes which can be accessed in parallel from the processing elements. The Vesta parallel file system is designed to exploit this architecture. Files are distributed across the I/O nodes, exposing an opportunity for high-bandwidth data transfer across the multicomputer's low-latency network. The Vesta interface supports a user-defined parallel view of file data, which gives users some control over the layout of data. This is useful for tailoring data layout to match common access patterns. The interface also allows user-defined partitioning and repartitioning of files without moving data among storage nodes.

The partitioning and control over layout are implemented without compromising scalability or parallelism. In fact, all data accesses are done directly to the I/O node that contains the requested data, without any indirection or access to shared metadata. Metadata describing the layout parameters is obtained once by each PE, when the file is first attached. Thereafter, each PE can compute the location of data directly and independently. The metadata itself is also distributed across the I/O nodes and is located by hashing, with no indirections.