Indicators for Self-Diagnosis: Communication-Based Performance Measures

Abstract

Multiagent systems (MAS) have found their way into industrial applications in recent years and appear to be one of the most promising technologies that originated in AI research in recent years. However, evaluation standards as they are common e.g. in the scheduling or database systems communities are largely amiss. In this paper, we propose communication-based performance measurement (CBPM) as a new method that is particularly suitable for open, communication-intensive MAS, and argue that it can be used as to design indicators for self-diagnosis by the MAS itself. The ability of such self-diagnosis is a prerequisite for MAS with self-repairing and self-optimising properties required by the autonomic computing view. CBPM is based on the idea that important aspects of the external behaviour of a MAS can be measured in terms of the communication processes within them. We present different levels of communication-based performance measurement: frequency analysis of performatives and analysis of complex message patterns. Several examples of analyses of inter-agent communication based on FIPA-ACL and the contract-net protocol in implemented, complex, market-oriented MAS demonstrate the usefulness of our approach. We conclude that these performance measures provide useful information about MAS and pave the way for devising autonomic self-improvement methods for these systems.