Accuracy of measurement techniques supporting QoS in packet-based intranet and extranet VPNs

Where VPNs are supported by packet based networks they will be monitored using both active and passive measurement techniques. A passive monitoring scheme that offers very significant levels of accuracy and stability, without traffic overhead, and regardless of the network utilisation, is developed and analysed. Where a VPN is carried over a number of different packet networks active measurement (active probing) is needed for performance monitoring. By analysis, validated using simulation, the effect of using active probing, in terms of the additional load implied by the desired level of accuracy, using typical models of packet traffic is investigated. Active packet probing is found to begin to require too much of the link bandwidth, at the desired level of accuracy, somewhere between a load of 60% and 80%, depending on the inherent burstiness of the traffic, and whether the probing packets have to determine just the mean delay, or the delay distribution (jitter) too. The increase in probing load with increasing network utilisation is highly nonlinear; while returning accurate results for moderate loads, under higher loads a sudden increase occurs, requiring so much bandwidth that this form of monitoring would become highly inaccurate.