Distributed Transaction Processing and The Camelot System

This paper describes distributed transaction processing, a technique used for simplifying the construction of reliable distributed systems. After introducing transaction processing, the paper presents models describing the structure of distributed systems, the transactional computations on them, and the layered software architecture that supports those computations. The software architecture model contains five layers, including an intermediate layer that provides a common set of useful functions for supporting the highly reliable operation of system services, such as data management, file management, and mail. The functions of this layer can be realized in what is termed a distributed transaction facility. The paper then describes one such facility — Camelot. Camelot provides flexible and high performance commit supervision, disk management, and recovery mechanisms that are useful for implementing a wide class of abstract data types, including large databases. It runs on the Unix-compatible Mach operating system and uses the standard Arpanet IP communication protocols. Presently, Camelot runs on RT PC’s and Vaxes, but it should also run on other computers including shared-memory multiprocessors.