Warehousing communication network monitoring data

The primitive network performance data collected by intrusively monitoring a communication network consists of details of test packets transmitted and received over the network. The delay experienced by these test packets and the number of test packets lost or duplicated during communication provide an indication of the network´s performance. However, the primitive network performance data need an enormous amount of processing to generate suitable performance summaries. The paper presents an approach of preprocessing the primitive network performance data whereby frequency distributions of delays experienced by test packets transmitted during a specific time interval are generated. Frequency distributions are reusable as they can provide a variety of summaries (for, for example, averages, variances, percentiles etc.). Moreover, frequency distributions derived at finer levels of granularity can be simply combined to generate frequency distributions at coarser levels of granularity. The ability to reuse processed data yields significant efficiency improvements for an information processing system