High-performance IPv6 address lookup in GPU-accelerated software routers

High-performance IPv6 address lookup in GPU-accelerated software routers
复制标题

DOI:
10.1016/j.jnca.2016.08.004
复制
发表时间:
2016-10-01
影响因子:
8.7
通讯作者:
Yao, Xin
Yao, Xin
中科院分区:
计算机科学2区
文献类型:
--
作者:
Lin, Feng;Wang, Gang;Yao, Xin

文献摘要

被引文献

相似文献

由于物理链路速度、路由表大小和互联网流量的不断增加,现代路由器已成为处理高吞吐量数据包的主要瓶颈。作为路由器最耗时的任务,设计高效的IPv6 IP查找方案面临着新的挑战。本文设计了IPv6下的并行布隆过滤器,并在GPU上实现,开发了一种新的GPU加速软件路由器GRv6。此外,我们还设计了两种支持动态前缀更新的方案,即,动态前缀插入方案和动态前缀删除方案。为了评估GRv6的性能,我们用NVIDIA GeForce GTX 580实现了它,并利用5个真实的IPv6路由表来证明IP查找引擎可以实现每秒3000次更新的静态路由表60 Gbps和动态路由表40 Gbps。(C)2016爱思唯尔有限公司版权所有。
Due to the ever-increasing physical link speed, routing table size and internet traffic, modern routers have been the major bottleneck to process packages with a high throughout. As the most time-consumption task of routers, designing efficient IP lookup schemes for IPv6 face new challenges. In this paper, we design parallel bloom filter for IPv6, and implement it on Graphics Processing Unit to develop a novel GPU-accelerated software router, GRv6. Moreover, we design two schemes to support dynamic prefix update, i.e., dynamic prefix insert scheme and dynamic prefix delete scheme. To evaluate the performance of GRv6, we implement it with NVIDIA GeForce GTX 580 and utilize 5 real-life IPv6 routing tables to demonstrate that the IP lookup engine could achieve 60 Gbps for static routing tables, and 40 Gbps for dynamic routing tables with 3000 updates per second. (C) 2016 Elsevier Ltd. All rights reserved.