Browse Books

Reviewer: John S. Edwards

Several parts of this work, which grew from the author's Ph.D. thesis submitted to Vrije University, have been published in such places as the ACM Computing Surveys, Proceedings of the AFIPS National Computer Conference 1987, Proceedings of the IEEE Computer Society 1988 International Conference on Computer Languages, and Proceedings of the USENIX Workshop on Experiences with Building Distributed and Multiprocessor Systems. The book's basic purpose is to relate the results of Bal's research, which defines a new model for implementing applications of distributed systems called the shared data-object model. The book fulfills this purpose and also describes the entire field of distributed programming, including discussions of many languages. The major goal of the book is to ease the implementation of distributed systems. Whether Orca will ever be widely used is secondary to the insights into distributed problem solving. As the state of the art moves forward, the capabilities of programs for distributed systems will continue to evolve. To date, most of the work has been on specialized networks of processors for (usually) real-time distributed applications such as telecommunications switching systems. In these machines, most applications are ad hoc and proprietary. This book represents a contribution to the field. The book begins by defining a logically distributed software system as “consist[ing] of multiple software processes that communicate by explicit message passing.” This definition excludes systems in which the software processes communicate through shared data. The work covers combinations of logically and physically distributed systems. The author describes the troubles encountered when the distributed system was based on an existing client-server operating system called Amoeba, also developed at Vrije University. While Amoeba was readily available and was used to support early investigations, the author discovered several limitations while attempting to develop a true distributed system capability. In an attempt to determine the state of the art, the discussion turns to several existing languages for programming distributed systems. Bal examines them with six objectives in mind: expressiveness, simple semantics, type security, efficiency, readability, and portability. He considered all the sample languages with respect to a general classification schema where the address space was either shared or not. The distributed address space category was further divided among several message communication styles, such as synchronous or asynchronous message passing. The shared address space category was divided into parallel functional languages, parallel logic languages, and languages with distributed data structures. Each of the languages examined had shortcomings when evaluated against the predetermined goals. Some, like Ada, were not designed with parallel processes in mind, and many were designed for implementing distributed services rather than for general distributed applications programming. Finally, “existing languages for distributed programming also have other shortcomings, like poor integration of sequential and distributed constructs, high complexity, and lack of type security. Although these issues are not directly related to distributed programming, they are important and need more attention.” With these observations in mind, the author embarks on the main portion of the work—the design of a new language, Orca, based on the shared data model. This model hides the physical communication from the user. Instead, the user can allow different parts of the program to communicate through logically shared data objects. The idea behind the shared data object is to allow the language implementation to manage the actual physical replication of data among the various processors, while the programmer is unaware of the replication. Fault tolerance is also handled at a level transparent to the user and is not discussed directly. “Orca is a procedural, strongly typed language. Its statements and expressions are fairly conventional…it supports records, unions, dynamic arrays, ser bags, and general graphs.” The remainder of the book is devoted to implementation details and examples of Orca programs. The book includes an extensive bibliography, an adequate index, and a set of short appendices giving several examples of classic programs in Orca. It seems to be intended to record a research direction and point toward further effort. The book was pre pared with computer support and thus the typography is flawless.