Characterizing bandwidth requirements of multiplexed connections with quality-of-service constraints

Commonly proposed approaches for computing bandwidth requirements of multiplexed connections with quality-of-service (QOS) guarantees rely on "effective bandwidth". Since the performance criterion used in these approaches characterizes only QOS requirements at a single multiplexing point, they cannot capture the impact of interactions in the network on bandwidth requirements for end-to-end QOS guarantees. Due to the computational complexity, these approaches are valid only in an asymptotic regime. This results in computation of "effective bandwidth" for each application based on the most stringent QOS requirement among all services supported, and may in fact lead to non-statistical multiplexing for a large portion of multiplexed connections. "Effective bandwidth" of multiplexed connections is typically the sum of "effective bandwidth" of individual connections. This linearity explicitly indicates a tendency of ignoring statistical multiplexing. This paper presents a new bandwidth computation technique. We construct bandwidth functions that cast reliably characterize bandwidth requirements of statistically multiplexed connections under the end-to-end QOS constraints in a network setting, and validate this property with simulation. This technique ensures bandwidth gains for service classes with variable bit rate sources. Based on bandwidth functions, we formulate a criterion for statistical multiplexing under the QOS constraints. The computation is simple, so this technique is suitable for real-time traffic control.<
>