A Statistical Mechanics of Distributed Resource Sharing Mechanisms

The problem of analyzing the performance of large-scale interacting distributed resource sharing mechanisms arises in packet broadcast networks as well as multiprocessor switching mechanisms, VLSI chip communications and distributed databases. Queueing theory has major limitations in analyzing such systems: it cannot handle tight interactions and dependencies and it requires a fine-grained analysis of the detailed dynamics of a system to obtain coarse-grained equilibrium results. This paper proposes an alternative approach to the problem based upon statistical mechanics. Using this approach it is possible to analyze the performance of intricate distributed resource sharing mechanisms with a relative ease. Moreover, the analogy to physical phenomena offers new performance measures (e.g., communication "energy" and "pressure") and physical insights to the behavior of such systems.