Kami : an anarchic approach to distributed computing

Abstract

This thesis presents a distributed computing system, Kami, which provides support for applications running in an environment of heterogeneous workstations linked together by a high speed network. It enables users to easily create distributed applications by providing a backbone infrastructure of localized daemons which operate in a peer-to-peer networking environment, providing support for software distribution, network communication, and data streaming suitable for use by coarse grained distributed applications. As a collective entity, kami daemons, each individually run on a single machine, form a cooperating anarchy of processes. These support their applications using adaptive algorithms with no form of centralized control. Instead of attempting to provide a controlled environment, this thesis assumes a heterogeneous and uncontrolled environment, and presents a model for distributed computation that is completely decentralized and uses multicast communication between workstations to form an ecology of co-operating processes, which actively attempt to maintain an equilibrium between the demands of their users and the capabilities of the workstations on which they are running.