Research Of The Implementation Mechanism Of Terabit High Performance Router

With development of service diversity of the IP network and rapid bandwidth growth of the optical transmission network, the backbone network has been evolving from the Gigabit level to the Terabit level. Due to the technology development disproportion of optical fiber transmission and electronics, the core router as network node has become the main bottleneck in the progress of modern networks. Thus developing high-performance routers with terabit switching capacity and multiple QoS services support has become the focus of studies on modern network technology.Aiming at the QoS implementation problems of the terabit high-performance router and according to the requirements on one of key projects of the National 863 Program for the Tenth Five-year Plan "The basic platform and experiment system for terabit scalable high performance IPv4/v6 routers", this thesis presents a preliminary analysis of the influencing factors for QoS performance of the network node such as delay and delay jitter, based on network calculus theory. We also attempt to find out a method to analyze costs of network-call for various QoS technologies, based on the analysis of different QoS technologies. According to the variable-length packet switching architecture applied in the project, we introduce the distributed idea and propose two novel scheduling algorithms: sequence distributed modified compensating round robin (SDMCRR) and parallel distributed modified compensating round robin (PDMCRR). In addition we study the queue management mechanism requirements of DiffServ architecture and thus propose a buffer management scheme: multiple thresholds random early detection (MTRED). Finally, according to the implementation requirement of QoS in the terabit router, we present a two-stage QoS scheduling architecture including the VOQ queue management based on output ports in the input side and priority based queue in the output side. The implementation scheme of this architecture in the terabit router is also given.The key contributions and innovations of this thesis include: An IP QoS classification method based upon fields is presented, which can analyze the costs of various QoS technologies and provide a theoretical instruction of feasibility analysis; Analysis models for network nodes' QoS performance in IntServ and DiffServ reference service architecture are proposed based on network calculus theory; A novel scheduling algorithms supporting variable-length packets and distributed implementation, named as sequence distributed modified compensating round robin(SDMCRR), is proposed;? The bidirectional rotating round robin SDMCRR scheduling structure is presented to improve throughput under asymmetric traffic loads, due to the input traffic asymmetry of network nodes;? A parallel distributed modified compensating round robin (PDMCRR) scheduling algorithm is proposed by improving SDMCRR on input ports parallel and per-port pipeline, according to the multiple ports switching in the terabit router. The analysis shows that PDMCRR meets the requirement and can schedule packets with at least 40 byte length on lOGbps ports in the terabit router.? The engineering implementation architecture for PDMCCR scheduling algorithm is presented. Detailed analysis and design of key parts are also given.? The time complexity and fairness of PDMCRR is analyzed, which is verified by the comparison results in the NS simulation;? A priority based buffer management mechanism named Multi-Thresholds Random Early Detection (MTRED) is proposed, which is simulated by NS and can well support EF and AF services in DiffServ;? A two-stage QoS scheduling architecture is proposed, including the VOQ queue management based on output ports in the input side and the priority based queue in the output side. The implementation scheme of this architecture in the terabit router is also presented.