Wireless Sensor Networking: The Pleasure and the Pain

Abstract

The demand for highly lightweight decentralised self-management of Wireless Sensor Networks (WSNs) has lead to the pursuit of emergent and bio-inspired solutions. I will introduce the WSN field briefly and highlight the aspects that differentiate it from 'normal' computing. I then present some of the research we have been doing in this field for decentralised network control and emergent systems management. Many of the algorithms produced to manage a WSN focus on one managerial aspect or parameter, limiting their usefulness and consuming already scarce resources. We have identified sets of common structures and elements of many well-known emergent algorithms. I present examples that exploit this to efficiently manage more than one managerial parameter or aspect. However, I also show how established evaluation methodologies are extremely misleading as when implementing the systems on actual devices we soon find some very unexpected results. I discuss this phenomenon, suggest causes and make some suggestions regarding the engineering of WSNs.