核心路由器中业务引擎交换吞吐量研究  

Research on Switching Throughput of Traffic Manager in Core Routers

作　　者：孙长华[1] 刘斌[1] 李文杰[1] 

机构地区：[1]清华大学计算机科学与技术系,北京100084

出　　处：《电子学报》2005年第7期1242-1246,共5页Acta Electronica Sinica

基　　金：国家863高科技研究发展计划(No.2002AA103011-1;No.2003AA115110);国家自然科学基金(No.60173009;No.60373007);中国-爱尔兰政府间国际科技合作项目(No.CI-2003-02)

摘　　要：本文给出了核心路由器中业务引擎吞吐量分析的一般模型,并应用此模型研究了实际设计的用于OC-48c接口的业务引擎.由此得出:当采用GigaStream交换芯片时,每路CSIX(CommonSwitchInterface)接口至少需用4路HSSL(HighSpeedSerialLink)链路;在CSIX接口处形成CFrame帧时需要对IP包最后一个信元按照实际大小形成CFrame帧,才能保证交换网络与CSIX接口处线速转发所有长度的IP包.最后,本文给出了业务引擎吞吐量和IP包存储之间的一般性关系,该关系可用于指导核心路由器交换网络的设计.We made a general model to analyze switching throughput of traffic manager in core reuters. By designing a real traffic manager which uses the OC-48c interface, we analyze the whole system and point out that at least four HSSLs should be employed per C.SIX interface when using Vitesse＇ s GigaStream switch chip set. Meamwhile, at the CSIX interface, the CFrame should be constructed according to the actual size of the last cell of each IP packet. Above principles can guarantee forwarding IP packets at line rate. We give a general relationship between throughput and buffeting scheme of IP packets in the external memory, which is useful in the design of switch fabric in core routers.

关 键 词：吞吐量 GigaStream交换芯片 CSIX接口 业务引擎 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]