Improving Performance Bounds in Feed-Forward Networks by Paying Multiplexing Only Once

Abstract:
Bounding performance characteristics in communication networks is an important and interesting issue. In this study we assume uncertainty about the way different flows in a network are multiplexed, we even drop the common FIFO assumption. Under so-called arbitrary multiplexing we derive new bounds for the tractable, yet non-trivial case of feed-forward networks. This is accomplished for pragmatic, but general traffic and server models using network calculus. In particular, we derive an end-to-end service curve for a flow of interest under arbitrary multiplexing, establishing what we call the pay multiplexing only once principle. Finally, we present some numerical results to compare our bounds against the best known bounds for networks of arbitrary multiplexing nodes.