Case Studies of Node-level QoS Adaptation in Cougaar

Multi-agent systems, like other software systems, have a set of abstract resource requirements for proper execution. These requirements, often known as Quality Of Service (QoS), change dynamically and can come into conflict with the availability of physical resources. The ability to adapt QoS requirements to changes in the environment is a key survivability feature in distributed multi-agent systems (MAS). For example, network and CPU resources are heavily stressed in operational environments, and systems that cannot adapt their requirements dynamically will be unable to function effectively. The Cougaar architecture supports successful QoS-adaptation by providing (1) a wide range of adaptation strategies, (2) a reliable characterization of the cost-benefit tradeoff for any given strategy, and (3) a high-level, accurate and timely data model of the dynamic resource constraints and application requirements. With this support available, a society of software agents can pick an effective strategy for adapting to whatever resource-limited situation it finds itself in, and therefore remain operational. This paper presents examples of the use of Cougaar mechanisms and node-level services that improve QoS-adaptation, along with discussions of these various techniques, both runtime and post-mortem, for evaluating the overhead these mechanisms impose. It also presents a detailed description of a suite of test societies, used in our examples, that has proven to be very useful for benchmarking Cougaar and evaluating adaptation options.