Performance Evaluation of Mobile Agents: Issues and Approaches

Abstract

With the emergence of Internet as a world-wide infrastructure for communication and information exchange, Internet-based distributed applications have gained remarkable popularity. One of the most promising approaches for developing such applications is the Java-based mobile-agent paradigm [5,8,15,39]. Mobile Agents (MA) are being used already in a variety of applications ranging from telecommunications management to Web databases, cooperative environments information-gathering systems, electronic commerce systems, intelligent distributed systems, and so on [23,24,7,9,4]. In that context, a distributed application can be considered as a dynamic group of agents working in coordination to accomplish some goal.

Mobile agents offer a number of diverse advantages in the development of distributed systems including enhanced programmability through modular, object-oriented structures and the increased flexibility provided by mobility; performance optimization for distributed operations that involve heavy network delays and/or weak connectivity; extended autonomy in terms of existing support for asynchronous execution and disconnected operations [6,8,21]. The employment of MA technologies for the development of next- generation Internet systems opens numerous research problems related to programming APIs and tools, security, fault-tolerance, design paradigms and programming techniques, communication, intelligence, scalability and performance [18].

The issue of performance, in particular, is very important in emerging Internet-systems: numerous studies show that performance of systems and applications determines to a large extent the popularity of Internet services and user-perceived Quality of Service [1,3]. Moreover, performance evaluation is crucial for performance “debugging,” that is the thorough understanding of performance behavior of systems. Results from performance analyses can enhance the discovery of performance and scalability bottlenecks, the quantitative comparison of different platforms and systems, the optimization of application designs, and the extrapolation of properties of future systems.

So far, systematic performance studies and experiments have provided important insights for the design of parallel and distributed systems and the tuning of applications [13,14,17,29]. Nevertheless, the more complex a system or application is, the harder its evaluation becomes, because of the large variety of factors that affect its performance [10]. This observation is particularly true for systems and applications running on top of gigantic platforms like Internet [1]. The performance evaluation of mobile-agent systems is even harder than the analysis of more traditional parallel and distributed systems, due to the dynamic nature and agile configuration of mobile agents.

The objective of this chapter is to investigate issues pertinent to performance engineering for mobile-agent platforms and systems. First, we describe briefly the basic characteristics of mobile agents. Then, we explore the notion of performance evaluation in the context of mobile agents and present an overview of recent approaches for performance evaluation and analysis of mobile-agent systems. These approaches include benchmarking efforts, scalability studies, analytical models, and Petri-net modeling.

Finally, we present a novel performance analysis approach that we developed to gauge quantitatively the performance characteristics of different mobile-agent platforms [12,29]. We materialize this approach as a hierarchical framework of benchmarks designed to isolate performance properties of interest, at different levels of detail [12,29]. We identify the structure and parameters of benchmarks and propose metrics that can be used to capture their properties. We present a set of proposed benchmarks and examine their behavior when implemented with commercial, Java-based, mobile-agent platforms.