A recoverable object store

Abstract

A design is presented for the storage component of a self-recovering distributed operating system. This component consists of an object manager, which maintains objects on main memory and on the disk, and a recovery layer, which incorporates a collection of highly optimized algorithms based on optimistic recovery. With optimistic recovery it is possible for a machine (or collection of machines) to present a fault-free interface to programs running on it (or them), making all data appear to be persistent. The optimizations presented make it possible to do this at a cost no higher than that of transaction systems.<>