Energy-efficient Itinerary Planning for Mobile Agents in Wireless Sensor Networks † Min Chen, † Victor Leung, ‡ Shiwen Mao, * Taekyoung Kwon and § Ming Li † Dept. of Electrical & Computer Engineering, Univ. of British Columbia, V6T 1Z4, Canada (minchen,vleung@ece.ubc.ca) ‡ Dept. of Electrical & Computer Engineering, Auburn Univ., Auburn, AL 36849-5201, USA (smao@ieee.org) * School of Computer Science & Engineering, Seoul National University, Seoul, 151-744, Korea (tkkwon@snu.ac.kr) § Dept. of Computer Science, California State Univ., Fresno, CA 93740, USA (mingli@csufresno.edu) Abstract—Compared to conventional wireless sensor networks (WSNs) that are operated based on the client-server computing model, mobile agent (MA) systems provide new capabilities for energy-efficient data dissemination by flexibly planning its itinerary for facilitating agent based data collection and aggre- gation. It has been known that finding the optimal itinerary is NP-hard and is still an open area of research. In this paper, we consider the impact of both data aggregation and energy- efficiency in sensor networks itinerary selection, We propose an itinerary energy minimum for first-source-selection (IEMF) algo- rithm, as well as the itinerary energy minimum algorithm (IEMA), the iterative version of IEMF. Our simulation experiments show that IEMF provides higher energy efficiency and lower delay compared to existing solutions, and IEMA outperforms IEMF with some moderate increase in computation complexity. I. I NTRODUCTION The application-specific nature of tasking a wireless sensor network (WSN) requires that sensor nodes have various ca- pabilities for multiple applications. It would be impractical to store in the local memory of embedded sensors all the programs needed to run every possible application, due to the tight memory constraints. A mobile agent (MA) is a special kind of software that migrates among network nodes to carry out task(s) autonomously and intelligently in response to changing conditions in the network environment, in order to achieve the objectives of the agent dispatcher. The use of MAs to dynamically deploy new applications in WSNs, has been proven to be an effective method to address this challenge. Recently there has been a growing interest on the design, development, and deployment of MA systems for high-level inference and surveillance in WSNs [1]–[8]. In [1], the agent design in WSNs is decomposed into four components, i.e., architecture, itinerary planning, middleware system design and agent cooperation. Among the four components, itinerary planning determines the order of source nodes to be visited during agent migration, which has a significant impact on energy performance of the MA system. It has been shown that finding an optimal itinerary is NP-hard. Therefore, heuristic al- gorithms are generally used to compute competitive itineraries with a sub-optimal performance. In [2], two simple heuristics are proposed: (i) a local closest first (LCF) scheme that searches for the next node with the shortest distance to the current node, and (ii) a global closest first (GCF) scheme that searches for the next node closest to the dispatcher. These two schemes only consider the spatial distances between sensor nodes and thus, may not be energy efficient in many cases. A genetic algorithm (GA) [3] is proposed to exploit the global information of sensor detection signal levels and link power consumption. In GA, every node reports its status to the sink node, which may incur considerable control overhead. It is neither scalable to network size nor a lightweight solution that is suitable for sensor nodes constrained in energy supply. The original LCF, GCF [2] and GA schemes [3] are all based on the following two assumptions: (i) a cluster-based network architecture, where all nodes (e.g., sink and source nodes) can communicate with each other in one hop; (ii) high redundancy among the sensory data, which can be fused into a single data packet with a fixed size. This implies that a perfect aggregation model is used. These assumptions limit the scope of the existing schemes. In this paper, we focus on designing lightweight, energy effi- cient itinerary planning algorithms without making the above assumptions. We first propose an itinerary energy minimum selection for first-source-selection (IEMF) algorithm, which extends LCF by choosing the first source node to visit based on estimated communication cost. In IEMF, the impact of both data aggregation and energy efficiency are taken into account to obtain an energy-efficient itinerary. The scheme is quite general, in the sense that it adopts a universal aggregation model, which facilitates the support for a wide range of applications. In addition, IEMF does not rely on any specific network architecture and is suitable for multi-hop WSNs. We also observe that IEMF achieves energy efficient itineraries without incurring additional control overhead, as compared with existing lightweight approaches such as LCF and GCF. Furthermore, we propose the itinerary energy minimum al- gorithm (IEMA), which is an iterative version of IEMF. During each iteration, IEMA selects an optimal source node as the next source to visit among the remaining set of source nodes. We show that with more iterations, the suboptimal itinerary can be progressively improved, while the largest reduction in average delay and energy consumption are achieves after the first few iterations. We can thus trade off between energy efficiency and computational complexity based on specific application requirements. The remainder of the paper is organized as follows. The problem is stated in Section II. We present IEMF and IEMA in Section III. Our simulation studies are reported in Section IV. Section V concludes the paper. 1