Oasis: An agent-oriented programming language for heterogeneous distributed environment.

We investigated an approach to constructing distributed applications whereby computational processes are modeled as a distributed collection of autonomous cooperative agents. This approach extends the classical object-oriented computing paradigm to a distributed environment in a natural way. Agent-oriented programming, according to our formulation, has two important advantages. Firstly, it presents a higher-level abstraction for world-modeling. Secondly, it supports flexible-sharing concomitant with computational autonomy, both of which are desirable for large-scale heterogeneous distributed environment. The major contribution of our work is to lend programming language support to the agent-oriented paradigm of computing. We have integrated the notion of an agent into a programming language type system, where it is given first-class status. Our research elaborates upon the agent-oriented paradigm by providing a concrete realization in the form of a programming language called Oasis, which has been implemented on several platforms. This thesis describes the Oasis agent model and illustrates the Oasis programming language. Our practical experience with constructing the Oasis compiler and run-time system on several popular Unix machines are also discussed. We have introduced the idea of a decoupled compiler system and have demonstrated its efficacy through a concrete implementation. Using a delayed translation mechanism, the Oasis decoupled compiler system delivers Oasis applications to the end-user on heterogeneous distributed platforms in a time and space efficient manner, with minimal administrative efforts in code management. We showed that the Oasis compiler generates good code across both RISC and CISC platforms through competitive benchmarking with existing compiler systems, including C, CommonLisp, and Standard ML. In addition, we also showed that the performance of the Oasis garbage collector, which uses a novel tag-free scheme adapted from previous work, is significantly better than that of the CommonLisp and Standard ML systems. Furthermore, we had programmed a distributed branch-and-bound algorithm in Oasis to solve the well-known Traveling Salesman Problem on a network of RS/6000 workstations, and have obtained respectable speedup using up to seven processors in parallel.PhDApplied SciencesComputer scienceUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/128975/2/9308288.pd