Abstract
The research in computer architecture for agent-based parallel and distributed processing was prompted by the interest and the related ongoing work in agent-based computing. Agent-based computing best suits the Internet (a collection of computers interconnected by high speed communication links) in which every computer or host is an entity. Agents enable the user to search the internet offline as compared to online. The user is freed to attend to task requiring the user's intervention and the user only logs to the network to check the results returned by the agent. A number of agent-based systems have been implemented. The TACOMA, Telescript and WAVE are based on an interpreted engine (program code) running in the background, accepting and interpreting the user inputs. The Swarm system on the other hand, is based on compiled program code which is stored at each host. The agent navigate the network based on destination address (PM address in Swarm, another agent in TACOMA, a place in Telescript or a node address in WAVE). In the interpreted program based systems, the agent carries with it the code to be executed at the reached destination, while in the Swarm system, the agent carries the instruction pointer, pointing to the next instruction to be executed at the reached destination. HYDRA; A Parallel and Distributed Swarm computer architecture for agent processing is based on a single user Swarm computer architecture prototype developed and simulated on a single host by [Errico 1996]. Agents are self contained executable mobile processes that navigate a network of hosts using hosts' addresses, executing a task, and when the task terminates, the outcome of the task execution is sent to the user. Swarm consists of three modules: the AP (Access Point) through which the user can access the Swarm computer to run tasks, the RM (Routing Module), responsible for routing the agents in the Swarm computer, and the PM (Processing Module), which is the execution engine in the Swarm computer. A PM is the destination of an executing agent. The aim of the thesis was to simulate the Swarm computer architecture in a distributed environment, where the PMs are distributed among several hosts in the network. The distributed environment consists of a few or several hosts connected together by an interconnection network. Hosts on the network are grouped into domains, and the domains are interconnected together to form a network. Hosts in a domain are directly connected to every other host within that domain. Within a domain, a single host acts as a gateway host handling extra-domain agents. To enable me achieve my goal, I designed, implemented and tested a Communication Router (CR). The CR consists of Network Router (NR) and the Routing Module (RM). The CR has two main functions: 1. Provide connectivity among the hosts in the network. 2. Route agents between the hosts. To be able to connect host on the network and to route agents between them, I developed a connection algorithm, and a routing algorithm. The connection algorithm logically connects hosts in the same domain, builds a routing table at each host, and identifies a gateway host for each domain. The routing algorithm, using the routing table at each host ensures delivery of agents to their proper destinations, irrespective of the location of the Processing Module. Simulation tests were carried on the the HYDRA computer prototype, and the preliminary test results obtained do show that the connection, and routing algorithms were respectively able to connect the hosts, and to route agents between the hosts. And, that agents and the HYDRA computer architecture could be used to do some useful work. In this thesis I present the HYDRA computer architecture, the work done in the design, and implementation of the Network Router, Routing Module, Access Point, the connection algorithm, the routing table, the agent routing algorithm, the communication strategy adopted for routing messages in the network, the simulation tests carried out, and the test results obtained.