Donald Bren School of Information and Computer Sciences

MESSENGERS is a distributed system based on the principles of autonomous objects. It facilitates distributed parallel computing as cooperative work among autonomous objects called Messengers, which carry their own behavior in the form of a program. Messengers exhibit two abilities: navigational autonomy and dynamic composition. They create and navigate through a modifiable computational network, and coordinate dynamic function invocations as they visit new nodes. Hence, problems including nondeterminism can be solved in parallel by Messengers. This flexibility is realized by having an interpreter daemon running at each physical node. The MESSENGERS interpreter daemon is distinct from that of other autonomous-object-based systems in terms of its support for efficient parallel processing. It functions as a micro-kernel for autonomous objects rather than a simple interpreter. It supports inter-Messengers communication, synchronization, and virtual-time-based function scheduling. In this paper, we describe the architecture of MESSENGERS, focusing specifically on performance-oriented features, present its performance evaluation, and discuss optimal granularity of each object.