A cost-benefit framework for online management of a metacomputing system
Yair Amir, Baruch Awerbuch, R. Sean Borgstrom · 1998
Managing a large collection of networked machines, with a series of incoming jobs, requires that the jobs be assigned to machines wisely. A new approach to this problem is presented, inspired by economic principles: the Cost-Benefit Framework. This framework simplifies complex assignment and admission control decisions, and performs well in practice. We demonstrate this framework in the context of an Internet-wide market for computational services and verify its utility for a classic network of workstations. 1.1 Keywords Networks, resource allocation, markets. 2. Introduction Collections of networked machines are common in the modern world. Using each individual machine as a completely independent computer is obviously inefficient -- one machine could be working on a dozen jobs while the others sit idle. A metacomputing system is a set of networked machines that can pool their computational resources to avoid this problem. Each machine has several computational resources associated...