Optimising data processing in network performance monitoring systems

Recent literature on the performance monitoring of communication networks focuses on exploiting compressibility of the collected data and its management on high-performance hardware. An alternative approach is to use summary structures referred to as intermediate information which support the reuse of semiprocessed information to optimise the calculation of nondistributive and nonalgebraic network delay statistics. Different intermediate-information structures possess different performance and accuracy characteristics. By selecting an appropriate structure on the basis of the query and the state of the intermediate-information cache, we show that significant improvements in the processing cost of a database of network measurements can be achieved. An approach to the dynamic selection of such a structure is presented and results showing the subsequent improvement in processing speed are given.