We take the position that large-scale distributed systems are better understood, at all levels, when locality is taken into account. When communication and mobility are clearly separated, it is easier to design, understand, and implement goal-directed agent programs. We present the Spider model of agents to validate our position. Systems contain two kinds of entities: spiders which represent service providers, and arms, which represent goal-directed agents. Communication, however, takes place only between an arm and the spider at which it is currently located. We present both a formal description of the model using the ambient calculus, and a Java-based implementation.
F. Y. Huang, David B. Skillicorn