一种多级无缓存高阶路由器的设计与实现

发布时间：2018-04-03 16:20

本文选题：高性能计算　切入点：高阶路由器　出处：《计算机工程与科学》2017年02期

【摘要】：随着高性能网络规模的增加,高阶路由器结构设计成为高性能计算研究的重点和热点。使用高阶路由器,网络能实现更低的报文传输延迟、网络功耗和网络构建成本,同时高阶路由器的应用还可以提高网络可靠性。高性能路由器的阶数不断提高,仅靠扩展单级crossbar交换结构的阶数使路由器内部的连线资源急速增长,交叉开关的实现代价将不可接受,这就需要为高阶路由器设计新型的交换结构。近十年来,出现了以YARC为代表的经典结构化设计以及"network within a network"等新型设计方法,未来的研究重点是解决高阶路由器结构设计中遇到的缓存、仲裁和扩展性等各种问题。鉴于此,实现了一种多级无缓存高阶路由器,这种高阶路由器内部是一个多级Clos网络,每一级有相应的仲裁模块对请求进行调度,数据包缓存在输入/输出端口实现,除去这些缓冲区单元,该网络是无缓存的。最后通过BookSim模拟器进行了大量的性能测试,所设计的路由器能够正常工作,性能良好。
[Abstract]:With the increase of the scale of high performance network, the structure design of high-order router becomes the focus and hotspot of high-performance computing.Using high-order routers, the network can achieve lower packet transmission delay, network power consumption and network construction cost, and the application of high-order routers can also improve network reliability.The order of high performance router is increasing continuously. Only by extending the order of single-stage crossbar switch structure, the connection resource inside router increases rapidly, and the cost of implementing crossover switch will be unacceptable.This requires the design of a new switching architecture for higher-order routers.In recent ten years, classical structured design, represented by YARC, and new design methods such as "network within a network" have emerged. In the future, the focus of research is to solve the problems of cache, arbitration and expansibility encountered in the structural design of high-order routers.In view of this, a multi-level cache free high-order router is implemented. This kind of high-order router is a multi-level Clos network, each level has a corresponding arbitration module to schedule the request, and the data packet cache is implemented in the input / output port.Without these buffer units, the network is cache-free.Finally, a large number of performance tests are carried out through the BookSim simulator, and the designed router can work normally and has good performance.
【作者单位】：国防科学技术大学计算机学院;
【基金】：国家自然科学基金(61272482) 全国百篇优秀博士论文基金(201450) 国家863计划(2015AA01A301)
【分类号】：TP393.05

【相似文献】