Reliability assessment of multiple-agent cooperating systems

Multiple agent systems are being increasingly used in real-time applications, safety-critical systems, etc. Many of these applications have soft real-time constraints and imprecise correctness criteria for the output value. A principal characteristic of such systems is the tradeoff between using an efficient but less accurate strategy and a sophisticated but more time-consuming strategy. This paper develops methods for assessing the reliability of these systems in terms of the timeliness and accuracy of the output, to aid decision making. Two commonly used coordination structures, parallel and pipeline, are used. Some adaptive schemes are proposed, i.e., strategies can be selected dynamically by agents at the run-time according to the input encountered in order to enhance the reliability of the system. The applicability of this approach is demonstrated by a practical multimedia client-server example.