Network Performance Measurements: a Wavelet Analysis for Compression and Feature Extraction

Monitored network data allows operators to gainvaluable insight into the health and status of anetwork. Whilst such data is useful for real-timeanalysis, there is often a need to post-processhistorical network performance data. Storage of themonitored data then becomes a serious issue asnetwork monitoring activities generate significantquantities of data. The work in this thesis is motivated by the need ofmeasuring the performance of high-speed networks. Such networks produce large amounts of data over along period of time, making the storage of thisinformation practically inefficient. A possiblesolution to this problem is to use lossy compressionon an on-line system that intelligently compressescomputer network measurements while preserving thequality in important characteristics of the signal. This thesis contributes to the knowledge by examiningtwo threshold estimation techniques, two thresholdapplication techniques and the impact of window sizeon the lossy compression performance. In additioneight different wavelets were examined in terms ofcompression performance, energy preservation, scalingbehaviour, quality attributes and Long Range Dependence.