摘要
用户级通信允许应用程序直接访问网络接口,减小了通信操作的软件层开销。为了支持用户级通信,高效的虚拟地址到物理地址的变换起到关键作用。本文提出了基于地址变换表的地址变换机制,虚实地址变换都在网络接口控制器上完成,变换过程不需要操作系统的参与,并且无需失效处理。采用这种机制,我们实现了基于PCI-X面向集群系统的互连通信子系统CNI。实际测试获得了2.4μs的最小单边延迟和850MB/s的峰值带宽。
User-level communications alleviate the software overhead of the communication subsystem by allowing applications to access the network interface directly. For that purpose, efficient virtual-to-physical address translation is critical. In this paper, we propose an address- translation-table-based address translation scheme where every translation is done on the network interface controller without OS involvement and without miss handling. Adopting this mechanism, we implement our communication subsystem CNI oriented to cluster systems based on PCI-X. The experimental results shows the lowest one-way latency 2.4μs and the 850MB/s peak bandwidth.
出处
《计算机工程与科学》
CSCD
2008年第9期154-157,共4页
Computer Engineering & Science
关键词
地址变换
用户级通信
RDMA
address translation
user-level communications
RDMA
