On the Quality of Computer Network Measurements

Sammanfattning: Due to the complex diversity of contemporary Internet-services, computer network measurements have gained considerable interest during recent years. Since they supply network research, development and operations with data important for network traffic modelling, performance and trend analysis, etc. The quality of these measurements affect the results of these activities and thus the perception of the network and its services. This thesis contains a systematic investigation of computer network measurements and a comprehensive overview of factors influencing the quality of performance parameters obtained from computer network measurements. This is done using a novel network performance framework consisting of four modules: Generation, Measurement, Analysis and Visualization. These modules cover all major aspects controlling the quality of computer network measurements and thus the validity of all kinds of conclusions based on them. One major source of error is the timestamp accuracy obtained from measurement hardware and software. Therefore, a method is presented that estimates the timestamp accuracy obtained from measurement hardware and software. The method has been used to evaluate the timestamp accuracy of some commonly used hardware (Agilent J6800/J6830A and Endace DAG 3.5E) and software (Packet Capture Library). Furthermore, the influence of analysis on the quality of performance parameters is discussed. An example demonstrates how the quality of a performance metric (bitrate) is affected by different measurement tools and analysis methods. The thesis also contains performance evaluations of traffic generators, how accurately application-level measurements describe network behaviour, and of the quality of performance parameters obtained from PING and J-OWAMP. The major conclusion is that measurement systems and tools must be calibrated, verified and validated for the task of interest before using them for computer network measurements. A guideline is presented on how to obtain performance parameters at a desired quality level.