Performance Bounds in Feed−Forward Networks under Blind Multiplexing
Ivan Martinović · 2006
Bounding performance characteristics in communication networks is an important and interesting issue. In this study we assume uncertainty about the way di erent ows in a network are multiplexed, we even drop the common FIFO assumption. Under so-called blind multiplexing we derive new bounds for the tractable, yet non-trivial case of feed-forward networks. This is accomplished for pragmatic, but general tra c and server models using network calculus. In particular, we derive an end-to-end service curve for a ow of interest under blind multiplexing, establishing what we call the pay multiplexing only once principle. We specify the algorithms necessary to apply this result in a network of blind multiplexing nodes. Since these algorithms may have prohibitive computational costs, we present strategies to reduce the computational e ort in a controlled manner such that the quality of the bounds is a ected as little as possible. Finally we present some numerical results from a network calculus tool we developed and compare our bounds against the best known bounds for networks of blind multiplexing nodes.