ABSTRACT Performance is a persistent quality of any software systems. Software performance engineering (SPE) encompasses efforts to describe and improve performance of systems at the early stages of development of the system. Multi-Agent Systems (MAS) are composed of autonomous entities called agents which cooperates together to solve complex distributed problems. Whatever complex the system, the quality of the system is an important parameter to be addressed. In this paper we are proposing an algorithm for predicting the performance of softwfrfare systems using Artificial Neural Network (ANN) approach. The algorithm is a new attempt in performance engineering of MAS. We have used ANN models for size estimation of the software (representative workload) which is an important parameter for assessing the performance in early stages of software development. Another significant contribution is assessment of performance by considering the data gathered during feasibility study. The ANN models are trained and validated for different data sets. The algorithm is validated for static properties of the RETSINA architecture. A case study on MAS is considered and the results are obtained using the validated ANN model.
