In general, wireless sensor networks are used to fetch information on spatio-temporal characteristics of the observed physical world, spawning numerous unforeseen applications. Due to the special nature of the deployment environment and sensor nodes intrinsic instability, network failure happens unpredictably. Besides, a number of applications, such as ecological habitat monitoring and accident detection, inherently rely on persistent and instantaneous sensing data. Therefore, network diagnosis, a process of deducing the exact root cause of a failure from a set of observed failure indications, becomes of great importance in the development of wireless sensor networks.

Based on two real world environment monitoring sensor network projects GreenOrbs and CitySee, this proposal add...[ Read more ]

In general, wireless sensor networks are used to fetch information on spatio-temporal characteristics of the observed physical world, spawning numerous unforeseen applications. Due to the special nature of the deployment environment and sensor nodes intrinsic instability, network failure happens unpredictably. Besides, a number of applications, such as ecological habitat monitoring and accident detection, inherently rely on persistent and instantaneous sensing data. Therefore, network diagnosis, a process of deducing the exact root cause of a failure from a set of observed failure indications, becomes of great importance in the development of wireless sensor networks.

Based on two real world environment monitoring sensor network projects GreenOrbs and CitySee, this proposal addresses three key aspects for network diagnosis and management, i.e., evidence collection, faulty link detection and network bottleneck detection. Through testbed evaluation, intensive simulations and real world implementations, I evaluate the performance of the proposed approaches and verify the applicability. Some of them are also applied to our system management tools for large-scale wireless sensor networks.