Capacity partitioning in the defragmented elastic single link

Abstract

The paper discusses blocking and throughput analytical models for a single link under dynamic multi-class traffic, assuming defragmentation is used to mitigate fragmentation losses. The models are shown to generate reversible Markov chains that impart scalability to the performance analysis, including under any truncation of the chain that preserves the transition rates between all surviving states. The notion of link capacity pseudo-partitioning is then introduced, showing that it may be used to provide fairness between two classes in an adaptive, softwarizable fashion