A Case Study of Organizational Effects in a Distributed Sensor Network
Bryan Horling, Roger Mailler, Victor Lesser · Adaptive Agents and Multi-Agents Systems · 2004
1. Overview Distributed sensor networks (DSN) as an example application of distributed situation assessment and more generally distributed resource allocation has been studied in the MAS community since its infancy. This environment is particularly interesting when investigating issues of scale, because practical scenarios can be envisioned employing sensor networks that are arbitrarily large both in number and geographic size, making purely centralized control inefficient. We propose using organizational structures as a primary means to address the challenges embodied by the DSN domain. Rather than employing a single organizational scheme, we have found that exploiting the strengths of a collection of heterogeneous organizational styles can be quite effective. The specific design employed by a system can have both dramatic and subtle consequences, making the agent’s organization interesting from both practical and research perspectives. We will show that the system’s performance can be greatly influenced by varying just one aspect of such an organization. The notion of “organizational design” is used in many different fields, and generally refers to how entities in a society act and relate with one another. A design’s objectives will depend on the desired solution characteristics, so for different problems one might specify organizations which aim towards scalability, reliability, speed, or efficiency, among other things. To date, relatively little work has been done in the multi-agent community analyzing the characteristics and tradeoffs of different organizational types. We will provide quantitative results of our design to address this. The organizational design used in this solution primarily attempts to address the scalability problem, by exploiting locality of reference and organizational constraints to impose limits on how far classes of both control and data messages propagate. The environment’s most limiting resource is a shared wireless communication medium, and we will therefore use this resource to describe the effects of the organization. Our design uses environmental partitioning to create localized regions of interaction. Within these partitions, agents take on particular and different roles which dictate their individual behaviors. We will see that the number