Search

Anomaly detection has been a field of intensive research over the last years. Along with that several works to evaluate anomaly detectors have been proposed. In this paper we argue four properties regarding ideal evaluation methodologies that cannot be answered by single current evaluation technique employed today. We therefore present an framework of an evaluation methodology that leverages traces from operational networks, simulation and emulation to satisfy the four properties.

Congestion is still a great problem which Internet faces. Locating congestion links is very important to execute traffic engineering and mitigate network congestion. Firstly,congestion localization model was presented. After introducing AS performance matrix model, a kind of congestion detection algorithm based on threshold was put forward. Combining AS performance matrix with topology information, congested link localization algorithm was put forward. Simulation results show the algorithm has higher detection rate and lower false positive rate under different topology sizes and congestion levels.

ISP network performance evaluation is very important to the improvement of application performance, the optimum competition of ISP network and the operation and management of Internet. ISP network consists of one or several autonomous systems (AS), so AS is chosen as a basic unit of performance evaluation. The performance measurement architecture of AS was put forward, and performance matrix model of AS was constructed. A method of AS performance evaluation was put forward based on AS performance matrix model. The evaluation method adopted weighted average of path performance between border routers as comprehensive performance metric of AS. Simulation results indicate that the method can evaluate AS performance effectively.