Automatic compilation of performance models for industrial Multi-Agent Systems

Industry 4.0 compliant Multi-Agent Systems (MAS) rely on a modular structure of applications. However, there is a lack of efficient methods to identify the performance limits of MAS, such as response time or system throughput. This methodological gap complicates the design of commercial MAS with strict performance requirements. This paper (i) proposes a method for automatic compilation of performance models from system logs and (ii) introduces methods based on Queueing Networks to identify the performance limits of a MAS. The outcome of the method is a semi-automatically assembled Queueing Network. It is a mathematical abstraction for modeling and simulation of system performance, which is consequently used by the system developer to validate that the MAS meets the performance requirements. The proposed method is demonstrated on a test-bed made of six minicomputers Raspberry Pi that host agents implemented in Jade Environment. This method brings the future modular industrial systems with emergent behavioral patterns nearer to broad acceptance by industrial enterprises.