|
Rate Control architecture can provide a feasible and effective traffic regulation so that guaranteed quality of services can be achieved for various applications with diverse traffic characteristics. In the thesis, we propose three service disciplines based on the architecture of rate control scheme. The Soft Rate-Controlled Virtual Clock (SRCVC) is a work-conserving service discipline. In SRCVC, each session is allocated a delay bias at every node along its path. The scheduler assigns each arriving packet a service tag, which is computed as the sum of the virtual time from a hypothetical rate controller and the delay bias. An SRCVC server always selects the packet with the shortest service tag for transmission. This scheme is an extension of Virtual Clock service discipline. However, by introducing the delay bias and soft rate controller, SRCVC avoids the coupling on delay bound and bandwidth allocation in Virtual Clock. Thus, the utilization and admission region can be significant improved when compared with Virtual Clock. Furthermore, we improve SRCVC to be a non-work-conserving service discipline called Hard Rate-Controlled Virtual Clock (HRCVC) to prevent the accumulation of burst size at downstream nodes. HRCVC uniformly distributes the allocation of buffer space inside networks and still provides the same upper delay bound as in SRCVC. Finally, we apply the rate control architecture to WFQ/WF2Q and develop a new non-work- conserving service discipline called Rate-Controlled Packet-by-packet Fair Queueing (RCPFQ). We show that the RCPFQ service discipline provides an upper bound on end-to- end delay for every leaky bucket constrained session. The delay bound can be calculated from its universal service curve and is tighter than the sum of local delay bounds at nodes on its path. RCPFQ outperforms WFQ and WF2Q in terms of freedom in resource allocation, buffer space requirement, end-to-end delay bound, end-to-end delay-jitter bound and the complexity of implementation for admission control.
|