Measurement and Prediction of Centrical/Peripheral Network Properties based on Regression Analysis - A Parametric Foundation for Performance Self-Management in WSNs (1308.3855v1)

Abstract: Predicting performance-related behavior of the underlying network structure becomes more and more indispensable in terms of the aspired application outcome quality. However, the reliable forecast of QoS metrics like packet transfer delay in wireless network systems is still a challenging task. Even though existing approaches are technically capable of determining such network properties under certain assumptions, they mostly abstract away from primal aspects that inherently have an essential impact on temporal network performance dynamics. Also, they usually require auxiliary resources to be implemented and deployed along with the actual network components. In the course of developing a lightweight measurement-based alternative for the self-inspection and prediction of volatile performance characteristics in environments of any kind, we selectively investigate the duration of message delivery and packet loss rate against various parameters peculiar to common radio network technologies like Wireless Sensor Networks (WSNs). Our hands-on experiments reveal the relations between the oftentimes underestimated medium access delay and a variety of main influencing factors including packet size, backoff period, and number of neighbor nodes contending for the communication medium. A closed formulation of selected weighted drivers facilitates the average-case prediction of inter-node packet transfer delays for arbitrary configurations of given network parameters even on resource-scarce WSN devices. We validate our prediction method against basic multi-hop networking scenarios. Yield field test results proof the basic feasibility and high precision of our approach to network property estimation in virtue of self-governed local measurements and regression-based calculations paving the way for a prospective self-management of network properties based upon autonomous distributed coordination.