On regulating open distributed systems.

摘要

We address a number of issues that arise in modern distributed systems, which consist of a large number of heterogeneous, semi-autonomous components, whose membership may change dynamically.; The first issue is communication in such distributed systems, under which, the communicators may not know the location, identity, or even the presence of those they communicate with. Several techniques have been developed to support such communication—which effectively decouples the producers of information from its consumers, both in time and in space. But the indefinite, and potentially global, reach of decoupled communication has a dark side, which may make a system using it less predictable, more brittle, and less safe. To alleviate such drawbacks of decoupled communication, we propose a mechanism that imposes decentralized regulation of its use.; The second issue is work in digital enterprises. There, the presence of goods, funds, and services is represented electronically, and their handling is carried out through acting upon such electronic representation. Moreover, it is increasingly common for electronic agents (e.g., stock-trading agents) to initiate such actions autonomously. Consequently, work in digital enterprises tends to be quite invisible, and carried out in enormous speed, making it difficult to control and to audit. We introduce a regulatory regime that combines enterprise-wide policies with flexible managerial controls. And we show the effectiveness of such controls over a wide spectrum of managerial styles.; The final issue we address is exceptions and faults that may occur in open distributed systems, including those deployed in digital enterprises. The problem here is different, and considerably more complex, than the problem of exceptions in programming languages, and it does not lend itself to conventional techniques of fault tolerance, which are concerned mostly with hardware faults.; Our treatment of all the above problems utilizes a scalable coordination and control mechanism for distributed systems, called law-governed interaction (LGI), as the underlying computational basis.