The processing and interpretation of communication network performance data
There is an increased demand for higher levels of network availability and reliability. Effective monitoring is necessary to help meet this demand. Loughborough University's High speed Network (HSN) group and many other research groups have preformed significant research related to the monitoring of communication networks and the subsequent processing of the information collected in a meaningful way. This thesis takes latency as an example performance metric. The term 'Data Exception' is then employed to describe delay data that is unusual or unexpected due to some fundamental change in the underlying network performance. Examples of such changes include significant changes in usage patterns or planned alterations. The objective of this work is to process and interpret such communication network performance data at higher levels of understanding, and will focus on three main points:- • Developing a rule based algorithm to automate the detection of Delay Data Exceptions. • Correlating Delay Data Exceptions in different routes in a network to detect the location and the characteristic of the event that caused these Exceptions. • Predicting the effect of an external event on network performance. In addition to the above three points, the research started by improving a previously published technique for detection and classification of Delay Data Exceptions. The nature of the delay patterns in a commercial communication network was the key issue in developing the algorithm for the first section of the work, and a Neural Network was used in the last two research areas. The monitored delay data used in this work was obtained from different sources; the historical performance data of a commercial network, data from simulation and monitoring of test network in previous related research, and also by monitoring two experimental test networks built in the laboratory. The results of the detection algorithm show an improvement in detection performance, and provide more generality and independency of the source of the delay data. The outputs of the approaches used in the event detection and the performance predictions work give good results, and show potentially the ability to locate the underlying events.