Burst level simulation of a queueing system with active-idle sources

Modeling traffic at coarse timescales is a common technique to deal with the scalability problem of high-speed network simulation. Accuracy is an important issue in such time-abstraction simulation. We aim to quantitatively understand accuracy degradation in the burst-level simulation. We identify queueing nonlinearity occurring at small timescales as a major contributor to the error. This explains why heavy traffic setting and traffic macrolevel burstiness favor the abstract simulation.