Priority-based Multiple Flow-Preemption for Load-Sharing on MPLS Networks

• Main Authors: Yung-chang Chen, 陳永昌
• Other Authors: none
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/26051749752795548128

碩士 === 國立中山大學 === 電機工程學系研究所 === 92 === MPLS, the next-generation backbone architecture, can speed up packet forwarding via label switching. However, if the traversed LSPs (Label Switching Paths) are in congestion, traffic may encounter serious throughput degradation due to packet loss. This performance degradation may become worse and worse unless another forwarding LSP for these traffic flows are allocated. Moreover, in this Thesis, we propose a preemption mechanism for higher-priority flows to obtain necessary resources (the bandwidth of a LSP) by preempting low-priority traffic flows. In the multiple-flow preemption mechanism, several LSPs between Ingress router and Egress router are established to forward traffic flows. Some of these traffic flows are higher priority with stringent QoS requirements. If no satisfactory throughput can be met, Egress router would have to feedback average throughput values to Ingress router, which in turn decides how many lower-priority traffic flows should be preempted and moved to another LSP. Consequently, Ingress router has to send bandwidth reservation messages to reserve bandwidth just released by those lower-priority flows. After that, every core router can process these MPLS packets with specified traffic class to meet their QoS requirements. Finally, for the purpose of demonstration, we embed multiple flow preemption modules into MNS simulator and use it to run some experiments. Since in our scheme, it is not necessary to reserve bandwidth for higher-priority traffic flows in advance, the overall bandwidth utilization can be increased. Furthermore, the higher-priority flows can meet their QoS requirements by preempting the lower-priority flows whenever it is needed.