期刊文献+

基于通用以太网卡的高性能通信库设计实现

Design and implementation high performance communication library based on commodity Ethernet NICs
下载PDF
导出
摘要 采用硬件无关的设计方法,在Linux下实现了一个基于通用以太网卡的高性能集群通信库HPCL/Ethernet。该库利用Linux操作系统的网络设备接口实现了对各种以太网NIC(Network Interface Card)的支持,在软件层实现一个轻量级通信协议支持消息的可靠传输。采用通道简化缓冲区管理,固定缓冲区策略来减少消息拷贝,采用中断回收技术、链路层直接处理消息来优化通信。测试结果表明,该通信库在传输小消息时具有很低延迟,在传输大消息时具有高带宽的特点。传输消息长度为1196bytes时带宽高达1.18E+08bytes/s,是千兆以太网卡硬件带宽的94.4%。 A high performance communication library independent of hardware called HPCL/Ethernet is implemented based on commodity Ethernet NICs on the Linux.HPCL/Ethernet supports commodity Ethernet NICs by using the network device interface on the Linux kernel.Message reliable delivery is guaranteed in the software layer.Channel abstract simplify buffer management and fixed-buffer policy adopted for avoiding message copy.Both interrupt reaping technology and the policy of message processing in link layer are also used to optimize communication.The experimental results show that the HPCL/Ethernet is characteristic of low latency with short payload message and high bandwidth with long payload message.It provides 1.18E+08 bytes/s of bandwidth which is 94.4% of hardware-level bandwidth for 1 196 bytes message.
出处 《计算机工程与应用》 CSCD 北大核心 2007年第12期112-115,共4页 Computer Engineering and Applications
基金 国家自然科学基金(the National Natural Science Foundation of China under Grant No.60373008)。
关键词 集群 通信库 通用以太网 中断回收技术 固定缓冲区策略 cluster communication library commodity ethernet interrupt reaping fixed-buffer policy
  • 相关文献

参考文献7

  • 1Chiola G,Ciaccio G.Efficient parallel processing on low-cost clusters with GAMMA active ports[J].Parallel Computing,2000-02.
  • 2Basu A,Buch V,Vogels W,et al.U-Net:A user-level network interface for parallel and distributed computing[C]//Proc of the Third International Symposium on High Performance Computer Architecture(HPCA),Feb 1997.
  • 3Shinji Sumimoto,Hiroshi Tezuka,Atsushi Hori.High performance communication using a commodity network for cluster systems[C]//the Ninth International Symposium on High Performance Distributed Computing(HPDC-9),IEEE,August 2000.
  • 4Piyush Shivam,Pete Wyckoff,Dhabaleswar Panda.EMP:Zero-copy OS-bypass NIC-driven gigabit ethernet message passing[C]//International Conference on Supercomputing 2001.ACM SIGARCH,2001.
  • 5The Message Passing Interface(MPI) Standard[S/OL].http://www.mpiforum.org.
  • 6Tezuka H,Hori A,Ishikawa Y,et al.PM:An operating system coordinated High performance communication library[C]//LNCS 1225.High-Performance Computing and Networking.[S.l.]:Springer-Verlag,1997:708-717.
  • 7The Public Netperf[EB/OL].http://www.netperf.org/.

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部