资源描述：
《面向片上网络延迟优化直通路径预构机制》由会员上传分享，免费在线阅读，更多相关内容在工程资料-天天文库。
1、面向片上网络延迟优化直通路径预构机制摘要：针对维序路由模式下报文传输时表现出的空间局部性，提出了一种适用于报文直向通信优化传输的直通路径预构(SFRP)路由器结构•通过在路由器内部输入端口N,S,E,W分别静态地预构一条连接其直向通信输出端口的开关直通路径，结合适当的重用以及中断机制，以期实现后续的报文在满足匹配条件的情况下可直接利用开关直通路径进行报文的传输，进而旁通开关分配流水站，降低报文传输延迟•合成通信负载下的模拟结果表明，SFRP路由器在报文注入率饱和前较BASE,BASE_LR,BASE_
2、LR_SPC和PSEUDO_CIRCUIT等四种路由器结构均可获得非常明显的性能提升，其性能提升最大可达到59%,46%,25.6%和9.5%.真实应用负载下的模拟结果表明，SFRP路由器与PSEUDO_CIRCUIT路由器的平均延迟相当,但较其他几种路由器结构可分别获得57%,45%和21%的性能提升.关键词：片上网络；维序路由；空间局部性；直通路径预构；路由器中图分类号：TP302.2文献标识码：AStraight-forwardingRoutePreconfigurationMechanismf
3、orLatencyOptimizationinNoCsTANGXian-tuol,WANGFengl,XINGZuo-chengl,WANGQing-linl,FUGui-taol,2(1.CollegeofComputer,NationalUnivofDefenseTechnology,Changsha,Hunan410073,China；2.BeijingSatelliteNavigationCenter,Beijing100094,China)Abstract：Weproposedastraig
4、ht—forwardingroutepre-configuration(SFRP)routerarchitectureforthecommunicationspatiallocalitywhenpacketstraverseunderdimension-orderedroutingmode,whichwasadaptedtothelatencyoptimizationforthepacketsstraightforwardingtraversaLInourSFRProute「acorrespondin
5、gstraight-forwardingroutewaspreconfiguredateachinputport，whichconnectedtheinputportwithitscorrespondingstraightforwardingoutputport.Combiningappropriateroutereusewithterminationmechanism,thesubsequentpacketssatisfyingthecomparativeconditionswereexpected
6、todirectlyforwardtocrossbarwithoutSAstage,hencereducingtheaveragelatencyforpacketstraversa1.OurevaluationwithsyntheticworkloadtrafficshowsthatSFRProutercanachieveobviousperformanceimprovementbyupto59%,46%,25.6%and9.5%respectivelybeforethepacketsinjectio
7、nrateissaturated,comparedwiththeBASE,BASE_LR,BASE_LR_SPCandPSEUDO_CIRCUITrouters.Intherealapplicationtrafficworkloads,theperformanceimprovementoftheSFRProuterisanalogoustothatofPSEUDO_CIRCUITrouter.Comparedwithotherthreekindsofrouters,SFRProutercanachie
8、veobviousperformanceimprovementbyupto57%,45%and21%respectively.Keywords:network-on-chip；dimension—orderedrouting；spatiallocality；straight—forwardingroutepre-configuration；routers半导体技术以及微处理器体系结构技术的飞速发展推动了芯片设计进入多核时代.随着单片上集成处理核规模的不断
