Stigmergic Learning for Self-Organizing Mobile Ad-Hoc Networks (MANET's)
H. Van Dyke Parunak, Sven Brueckner · Adaptive Agents and Multi-Agents Systems · 2004
Mobile ad-hoc networks (MANET’s) [1] present challenges that overwhelm traditional network management. MANET’s are highly dynamic, severely constrained in processing and communications resources, distributed and decentralized. Centralized management approaches requiring accurate and detailed knowledge about the state of the overall system may fail, while decentralized distributed strategies become competitive. This paper explores the applicability of fine-grained agent algorithms used in biological systems [7] to MANET’s. A network of moving nodes communicate within a limited range, and may fail temporarily. The nodes might be foot soldiers, or rovers exploring Mars, equipped with line-of-sight radios. Every node carries a client process. Some nodes also carry a server process (e.g., long-range comms links; wide-range sensors; specialized databases). Weight and space constraints limit the power available for communications and processing. Some services impose high power demands, so servers should operate only when needed to support demands from other nodes. Node movement must both achieve mission objectives and maintain communication connectivity. The server management problem answers three questions: given the current network topology determined by node locations, communications ranges and node availability, decide 1) which server nodes should actually expend battery power to execute the server process; 2) to which server node a particular client should send its next service request; and 3) where to relocate server nodes to meet current demand. MANET’s are an active area of current research, until recently focused on issues such as routing [5], access control [3], and security [11]. Recent research considers the service discovery problem, using service brokers to maintain directories of available servers [2, 4, 6, 10]. This research addresses the second question of the server management problem but not the first and third, and does not contemplate highly dynamic situations that can frustrate directories. Our approach solves all three problems and does not require centralized directories. 2. Emergent MANET Management