Fault tolerance in distributed UNIX

Abstract

An initial design for a fault tolerant, distributed version of UNIX was presented in an earlier paper [2]. That design left a number of open questions in two particular areas: Fault tolerance for server processes through which peripherals are accessed; recovery after a crash including the re-backup of processes. Since then, the fundamental design involving three-way message transmission has remained unchanged. However, server fault tolerance has been redesigned and is now more consistent with the fault tolerance of normal user processes. Recovery and re-backup have been completed in a more efficient manner than previously envisioned. In addition, important changes in the implementation have occurred. In this paper, we review the original design, borrowing heavily from the earlier paper in sections 1–3, and explain additions and modifications in later sections.