Distributed operating systems

Reviewer: John George Fletcher

This paper is a review of the current state of the art in distributed operating systems. It contrasts such a system, in which the “users . . . should not know (or care) on which machine . . .:9- Ttheir programs are running,” with the much less transparent notion of a network operating system. The authors then discuss various principles and techniques relating to the design of such systems, particularly in regard to communication primitives, naming and protection, resource management, fault tolerance, and services provided. It concludes with sketches of four actual distributed systems: the Cambridge distributed computing system, Amoeba, the V Kernel, and the Eden project. On the whole, the paper is accurate, complete, and clear. This reviewer recommends it highly, with the following reservations: (1)The various aspects of a distributed system are perhaps discussed a bit too independently of one another. It is not mentioned how certain concepts fit together and how some concepts solve (or cause) several problems at once. (2)Criticisms of several popular ideas are often too gentle, frequently only being implied. The inefficiency of the OSI design is noted, without mentioning how unnecessary its complexity is. Highly theoretical analyses, particularly in regard to scheduling, are discussed at length, with only a passing phrase or two about their impracticality. Much of the discussion is phrased so as to sound familiar and comforting to UNIX aficionados, until late in the paper a reference is made to “poor performance and a fair amount of effort spent trying to convince UNIX to do things against its will.”