PSL: Protocols and Pragmatics for Open Systems
Doug Lea and Jos Marlowe
Introduction by Doug Lea and Jos Marlowe
This paper is an expanded version of one appearing at the European Conference on Object Oriented Programming, 1995).
Our work was an outgrowth of the Sun Labs ADL project. The ADL group focussed on "lightweight" specification techniques, mainly for C++ programs. We wanted to produce similar frameworks covering the more challenging domain of typed distributed object systems -- initially targeting distributed CORBA applications. This was even harder than it looked!
The result, PSL, is a set of notations and rules for specifying and describing the behavior of CORBA applications. But the main enduring result of this project was to provide a deeper understanding of the basic nature of open distributed systems.
Specification languages and frameworks rarely become widely used in practice, and PSL was no exception. But in our case, the reasons probably have less to do with formalism itself, and more to do with the phenomena it addressed. PSL directly reflects the highly dynamic nature of open distributed systems, including the fact that the implementation of an interface need not always be the same, or even a single object, the use of asynchronous messages, failure, and dynamic reconfigurability. These and related issues can make it impractically difficult to say exactly what a given distributed system should or will do, even with the best notation and tools. (There are of course still contexts where it is very much worth attempting a full specification, in which case PSL provides methods and tools to do so.) At the time we were dealing with these issues, some of our conclusions about the nature and problems of such specifications were surprising to colleagues and readers. However, the paper's characterization of properties is surely less surprising today, given the prevalence of the Java(TM) Platform, which more directly supports open systems via dynamic loading, mobility etc., that were captured by PSL but were rare and only indirectly supported in CORBA systems during our period of study.
Formal characterization and specification of open distributed systems remain active areas of research. Interested readers might wish to read, for example, recent work on "Ambits" by Luca Cardelli and colleagues.