Research On The Adaptive Resource Preemption Strategy And The Preemption Algorithm In The DS-TE Networks

With the growing popularity of the Internet, the primary services are changing from the traditional file transfer, e-mail and Telnet to the multimedia application, which needs the network to provide integrated services with different QoS level (such as latency, bandwidth, packets losing rate guarantee). Thus, IETF introduced DiffServ-aware MPLS Traffic Engineering (DS-TE), which combines the good scalability of DiffServ and the effective routing strategy of MPLS traffic engineering and allows the regional network resources to be divided in accordance with the class of service. Thus the networks will be subdivided into virtual networks with different QoS capabilities.In the DS-TE networks, resource preemption strategy is an important strategy for bandwidth set and management, which ensures the priority business reliable high quality on the DiffServ differentiated service environment. And it will be more effective especially when the network load or network node or link failure occur. It had been proven that the problem of deciding the combination of connections for preemption is NP-complete in nature, so in the actual network a simple heuristic algorithm should be designed. In this paper, based on the analysis of preemption algorithm such as Min-Conn, Min-BW, V-PREPT and H-PREPT, a new preemption algorithm is proposed--Adaptive preemption algorithm whose basic thought is to equip LSR with the capability of managing the LSP bandwidth resources by strengthening the node LSR resource control over the LSP on the links. When the LSR hard preemption occurs, LSR chooses some lower priority LSP and reduces their transmission rate to adapt to the new LSP bandwidth demand. When the LSR rate reduction occurs on the initiating terminal, LDP will update each LSR bandwidth of the LSP. To avoid the parameters configuration, the adaptive preemption algorithm is divided into A-minconn algorithm and A-minpri algorithm. A-minconn algorithm lays emphasis on reducing the numbers of preempted LSP and bandwidth waste, while A-minpri algorithm lays emphasis on reducing preemption priority and bandwidth waste. When the preemption occurs, the two algorithms can be chosen flexibly in accordance with different emphases. Simulation results show that compared with the Min-Conn algorithm and V-PREPT algorithm, the adaptive preemption algorithm is suitable for the actual network, paying less cost, having more accurate preemption results and avoiding the parameters configuration for the cost function.