Real time detection and prediction of link level characteristics of computer networks

The focus of this thesis is on developing real time detection, estimation and prediction mechanisms for link level characteristics of computer networks, which can be utilized to enhance the stability, efficiency as well as security of the networks.;Early and accurate knowledge of link level characteristics in data communication networks is very useful for network managers to maintain the networks or for network designers to enhance the quality of service for users. To maintain the stability of the Internet despite of link failures, network managers wish to detect link failures as soon as they occur. In wireless networks, efficiency of energy consumption and data transmissions can be enhanced via adaptive data transmissions based on current and future wireless channel conditions. Wireless link quality detection and prediction tools are thus very useful for aiding the design of medium access control (MAC), routing and application layer protocols that adapt to changing link conditions. What's more, some abrupt changes in link level characteristics can be utilized to detect the intrusions and enhance the security of wireless networks.;The work presented in this thesis consists of three parts. In the first part, a Bayesian approach is proposed for time efficient link failure detection in interdomain. The detection is done using an automated mechanism to label, train and classify the network status based on features extracted from traces of Border Gateway Protocol (BGP) messages. We also provide tools for processing measurements that are fast enough to detect the anomalies in real time.;The second part of our work addresses the problem of long range prediction and early detection of link quality in wireless networks. We first construct an accurate, low-complexity mechanism for prediction of the future trend of the fading signal strength in large and medium time scales. Our method is independent of the propagation environment and distance between user nodes and the access points. We then capture both local details (small scale fading) as well as the global trend of the signal strength (medium and large scale fading), using multi-resolution wavelet analysis of the received signal strength. By expansion of the signal into wavelet basis, our method can predict the non-stationary received signal strength in realistic and fast varying wireless environments. To automatically obtain more knowledge about the propagation environments, we further propose a mechanism to detect the instances of a receiver entering or leaving shadow regions using received signal strength measurements. Since shadow fading can cause at least 6 dB power loss for 10% of the time, our early detection mechanism plays an important part in facilitating the design of adaptive data transmission schemes.;In the third part of this thesis, we propose a mechanism for detecting session hijacking attacks in wireless networks based on the analysis of the received signal strength. We first develop a model to describe the changes in the received signal strength of a wireless station during a session hijacking event while embedded in colored noise caused by fading wireless channels. An optimal filter is then designed to detect the abrupt changes caused by the attacks and distinguish them from those caused by small scale fading and shadow fading. The detection mechanism is validated using both simulation and experimental results.