Development, Analysis and Evaluation of Performance Models for Mobile Multi-Agent Networks

Abstract

Mobile agents have a number of interesting features such as creation, disposal of agents, execution of services at various network nodes, migration to other network nodes and communication with other agents. Such features are related to the performance area. Based on some examples, scenarios, or case studies, Petri net models (ordinary Petri net, colored Petri net or predicate/transition net), mathematical models, simulation models and queuing models have been developed in the literature. Such research works concentrate on studying the performance of mobile agent networks with much attention on deadlock problems, dynamic behavior problems and the calculation of response time problems. In order to better understand how to design distributed systems using the mobile agent paradigm and how to build a model with the capability of studying all such performance problems in an easy and a realistic way, we propose a new mobile agent performance model using the capability of the generalized stochastic Petri net (GSPN) modeling technique. We amended this model with new mobile agent behaviors to fully describe the dynamic behavior of the mobile agent network when it manipulates parallel/multiple agents, and uses some of the most important agent communications (e.g. remote, local, direct, indirect and parallel). Furthermore, the proposed model describes the creation process of new agents during the migration process, the execution of tasks among the network nodes, the way of handling the agent and its service task at each node and the interaction of a group of agents (as a collective result of the behaviors of each individual agent). To the best of our knowledge, such aspects are not included into one model because it is difficult to describe all these aspects into one model. Therefore, the developed GSPN mobile model gives the facility to study the performance of mobile agent networks with more details. In the performance analysis, we develop four studies to investigate the effect of different parameters such as agent communication time, size of mobile agent, number of mobile agents and number of hops on the performance of mobile agent networks. By performance of mobile agent networks, we mean the response time of the user's request. To the best of our knowledge, such four performance studies are not investigated previously by the GSPN modeling technique. Finally, we propose a reduction-modeling methodology to facilitate the modeling process of the practical mobile agent network that incorporates a large number of network nodes, a large number of users, and subsequently a large number of parallel/multiple agents. Such methodology reduces the original model to a simpler one while preserving the basic features and properties of the original model.