Fast, stable simulation of power-law packet traffic using concatenated acceleration techniques
In recent papers a traffic aggregation (TA) technique for power-law traffic has been investigated. It has been shown that TA provides very significant acceleration with highly stable results, when simulating packet network traffic with Markovian or power-law distributed activity periods. How the well-known rare-event simulation technique RESTART (repetitive simulation trials after reaching thresholds) can be applied in conjunction with TA is investigated. Applying for the first time two combined acceleration methods it is shown that this idea provides a technique that is robust, efficient and effective, allowing stable estimates of tail probabilities faster than with either individually. The results show that concatenating these acceleration techniques can provide speedup in excess of that provided by the sum of them individually. Because TA abridges the level of complexity involved in using RESTART, using them together is simpler than just applying RESTART to the same (original) network. This combination provides a powerful tool for tackling otherwise prohibitively time-consuming simulations of the sort of power-law governed packet traffic patterns frequently reported in the recent literature.
