Affiliations: Department of Computer Science, University of Greenwich, Old Royal Naval College, Park Row, Greenwich, London, SE10 9LS, UK. Tel.: +44 20 8331 8482; Fax: +44 20 8331 8665; E-mail: [email protected]
Abstract: This paper examines some of the emergence behaviours of natural distributed systems, with respect to identifying ways in which the highly successful strategies and mechanisms found in these systems can be employed in autonomic distributed computer systems. Particular emphasis is placed on communications aspects. The relative effectiveness of deterministic behaviour found in traditionally designed distributed applications and non-deterministic behaviour found in natural systems are explored. Ways in which deterministic design can lead to increases in complexity and conflicts between the non-functional requirements of distributed applications; that include scalability, efficiency, robustness, low-latency and stability; are identified. Deterministic design is contrasted with non-deterministic approaches, and ways in which the latter can avoid such conflicts and reduce the interaction and communication complexity are discussed. This paper proposes that we look to natural systems for inspiration when designing architecture and selecting communications strategies. To demonstrate the effectiveness of this paradigm, a nature-inspired election algorithm is presented. The design and communication strategy are described. The election algorithm encapsulates non-deterministic behaviour internally to achieve the non-functional requirements described above. Externally the algorithm exhibits deterministic behaviour in meeting its functional requirements. The performance and complexity are critically evaluated through a simulation model and an implementation.
