THE VESTA PARALLEL FILE SYSTEM

The Vests parallel file system is designed to provide parallel file ac cess to application programs running on multicomputers with parallel I /O subsystems. Vesta uses a new abstraction of files: a file is not a sequence of bytes, but rather it can be partitioned into multiple disj oint sequences that are accessed in parallel. The partitioning-which c an also be changed dynamically-reduces the need for synchronization an d coordination during the access. Some control over the layout of data is also provided, so the layout can be matched with the anticipated a ccess patterns. The system is fully implemented and forms the basis fo r the AM Parallel I/O File System on the IBM SP2. The implementation d oes not compromise scalability or parallelism. In fact, all data acces ses are done directly to the I/O node that contains the requested data , without any indirection or access to shared metadata. Disk mapping a nd caching functions are confined to each I/O node, so there is no nee d to keep data coherent across nodes. Performance measurements show go od scalability with increased resources. Moreover, different access pa tterns are shown to achieve similar performance.