Statistical Scaling Analysis of Localized Algorithms in Wireless Ad Hoc Sensor Networks

Farinaz Koushanfar and Miodrag Potkonjak1
(Professor Alberto L. Sangiovanni-Vincentelli)

Scaling can be defined as the quantitative characterization of properties, events, and mechanisms as the instance of the object under consideration increases its size. In many fields, from physics and computational biology to multiprocessor systems and the Internet, scaling is widely studied. Often, the key prerequisite for the application of an architecture, piece of system software, or an algorithm, is how it scales. Until now, scaling has not been systematically studied in the context of wireless ad hoc sensor networks.

From the design point of view, the key novelty is that our goal is not just to analyze how already existing algorithms and mechanisms scale, but also to develop insights and techniques for the development of scalable algorithms. From a methodological point of view, the main novelty is the use of statistical techniques for analysis of scaling, including non-parametric modeling and validation, sampling, correlation, and perturbation. Special emphasis is placed on localized algorithms. We introduce several generic techniques for the development of localized algorithms and study how they scale.

1Outside Adviser (non-EECS), UCLA

Send mail to the author : (farinaz@eecs.berkeley.edu)


Edit this abstract