Swarm Intelligence in Autonomic Computing: The Particle Swarm Optimization Case

Swarm Intelligence in Autonomic Computing: The Particle Swarm Optimization Case

Christos Anagnostopoulos (University of Athens, Greece) and Stathes Hadjiefthymiades (University of Athens, Greece)
DOI: 10.4018/978-1-60960-845-3.ch004

Abstract

Autonomic computing has become increasingly popular during recent years. Many mobile autonomic and context-aware applications exhibit self-organization in dynamic environments adopted from multi-agent, or swarm, research. The basic paradigm behind swarm systems is that tasks can be more efficiently dispatched through the use of multiple, simple autonomous agents instead of a single, sophisticated one. Such systems are much more adaptive, scalable, and robust than those based on a single, highly capable, agent. A swarm system can generally be defined as a decentralized group (swarm) of autonomous agents (particles) that are simple, with limited processing capabilities. Particles must cooperate intelligently to achieve common tasks.
Chapter Preview
Top

Introduction

One of the scientific areas of autonomic computing is the investigation of mechanisms that exploit the collaborative behavior of the agents in order to deal with the information discovery. Specifically, in discovering contextual information an agent (e.g., mobile node) needs to explore, locate and track the source that generates the required contextual information – context (e.g., environmental parameters like temperature, humidity, situations like fire outbreak) for the executing context-aware, mobile application (e.g., the control of a group of robots). On the other hand, Swarm Intelligence (SI) (including Ant Colony Optimization, Particle Swarm Optimization, and Stochastic Diffusion Search) introduces a powerful new paradigm for building fully distributed systems in which overall system functionality is attained by the interaction of individual agents with each other and with their environment. Such agents coordinate using decentralized control and self-organization. Swarm systems are intrinsically highly parallel and exhibit high levels of robustness and reliability.

In this chapter, we discuss the application of Particle Swarm Intelligence in autonomic computing and networking. We map the basic concepts and definitions of particle swarm intelligence to autonomic computing and investigate the contextual information exploration (detection, discovery and exploitation) in autonomous dynamic environments. Specifically we focus on those scenarios where autonomous nodes with autonomic and context aware applications are autonomously adjusting their knowledge, policies and learning models in trying to (even physically) locate up-to-date contextual information, captured by other nodes. We also establish the concept of contextual information quality (an ageing framework deprecates contextual information thus leading to low quality) and how the nodes autonomously attempt to discover context in a collaborative manner. Nodes with low quality context cannot capture such information by themselves but are in need for “fresh” context in order to feed their application. In addition, we discuss the introduction of particle swarm optimization process in contextual information discovery and we report the performance of the reported algorithms in this area through simulations. Finally, we report a set of exercises for experimenting on the discussed algorithms focusing on better understanding of the swarm intelligence framework in autonomic computing.

Swarm Intelligence

Swarm Intelligence (SI) introduces a powerful new paradigm for building fully distributed systems in which overall system functionality is attained by the interaction of individual agents with each other and with their environment. Such agents coordinate using decentralized control and self-organization. Swarm systems are intrinsically highly parallel and exhibit high levels of robustness and reliability:

  • A SI-driven distributed system does not have hierarchical command and control structure and thus no-single failure point or vulnerability. Agents are often very simple and the overall swarm is intrinsically fault-tolerant since it consists of a number of identical units operating (sensing context) and cooperating (sharing context) in parallel. In contrast, a conventional complex distributed system requires considerable design effort to achieve fault tolerance.

  • The key central concept in a swarm system is the simplicity of the agents -an agent can be a mobile phone carrying sensors. Simply increasing the number of agents assigned to a task (e.g., sensing context) does not necessarily improve the system’s performance (i.e., efficiency and reliability). Agents collaborate by exchanging useful information in order to obtain the required context.

  • In a totally distributed environment agents collaborate for discovering context with certain validity (e.g., related to time and/or space constraints). Context periodically turns obsolete and has to be regularly determined and discovered. Moreover, the resources of simple agents are limited in terms of

    • A.

      Memory: agents remember the history of their operation up to a certain extent,

    • B.

      Sensing Capabilities: for agents moving around, the sensing radius can be small enough relatively to the coverage area once possible neighboring agents can provide analogous local information, and

    • C.

      Communication Resources: communication among agents is intended solely to convey information on the swarm.

The above-mentioned points lead us to examine the adoption of the SI paradigm in context discovery.

Complete Chapter List

Search this Book:
Reset