The Episode File System

We describe the design of Episode, TM a highly portable POSIX-compliant le system. Episode is designed to utilize the disk bandwidth e ciently, and to scale well with improvements in disk capacity and speed. It utilizes logging of meta-data to obtain good performance, and to restart quickly after a crash. Episode uses a layered architecture and a generalization of les called containers to implement lesets. A leset is a logical le system representing a connected subtree. Filesets are the unit of administration, replication, and backup in Episode. The system works well, both as a standalone le system and as a distributed le system integrated with the OSF's Distributed Computing Environment (DCE). Episode will be shipped with the DCE as the Local File System component, and is also exportable by NFS. As for performance, Episode meta-data operations are signi cantly faster than typical UNIX Berkeley Fast File System implementations due to Episode's use of logging, while normal I/O operations run near disk capacity.