Kai-yuh Hsiao, Peter Gorniak, Deb Roy

Abstract

Any intelligent system, interacting with humans and acting in the real world, will be composed of many parts, each performing disparate and specialized tasks with wildly differing requirements. We have implemented a network abstraction layer, called NetP, that allows processes to communicate by sending structured data via non-blocking broadcasts to named channels. Any process can subscribe to any channel and receive the data posted to that channel. Using our API requires minimal programming effort, provides a simple network abstraction, and facilitates the debugging of network operations. The resulting system is resilient to partial failures and rapidly reconfigurable. In this paper, we present our networking approach, along with a case study that motivated our target feature list.