一种基于GPU的高效NDN数据名查找算法
首发时间:2017-04-28
摘要:命名数据网络(Named Data Network,NDN)是一种新型网络体系架构,它在转发数据包时是以数据名作为核心而不是IP地址。因此,数据名查找已经成为NDN节点的关键功能。然而,由于数据名的复杂,不定长等特性,使得高效数据名查找已经成为了一个大的挑战。在已有的研究工作中,有一种采用多对齐迁移数组(Multi-Aligned Transition Array, MATA)这一数据结构来压缩存储,并借助图形处理单元(Graphic Processing Units, GPU)的强大并行性来加速查找的方法很好的解决了该挑战。但是,由于MATA本身的存储效率较低,使得该方法在存储开销上仍然较大。因此,本文设计了一种新的数据结构:候选对齐迁移数组(Candidate-Aligned Transition Array, CATA)来解决存储开销问题。而且,实验结果表明CATA在存储利用率上达到了90%左右,同时存储开销相比MATA节省了80%,并且在查找性能上的表现上也与MATA基本相同。
For information in English, please click here
A GPU-based approach to achieve wire speed NDN name lookup with lower memory
Abstract:Named Data Networking (NDN) is a new network architecture which forwards packets according to their names rather than IP address. Therefore, name lookup becomes a key function of NDN nodes. However, due to the NDN name's unbounded number of components, name lookup brings many challenges which not only in the aspect of memory storage but also lookup speed. To solve these challenges, numerous researchers have contributed plenty of ideas, and a Graphic Processing Units (GPU)-based method which implemented multi-aligned transition array (MATA) structure has a good performance. But, because of the lower storage memory utilization, the MATA also wasted a lot of memory storage. In this paper, we focus on the utilization of memory, and put forward a candidate aligned transition array (CATA) structure to solve the problem of memory cost. Besides, we also deployed our method on the CPU-GPU engine, and the final experiment results show that our CATA structure can reduce almost 80 percent storage memory compares to MATA and the memory utilization is as higher as 90 percent. What's more, CATA also performs as well as MATA in the speed of lookup.
Keywords: Computer network NDN name lookup CATA
基金:
论文图表:
引用
No.4726438119305414****
同行评议
共计0人参与
勘误表
一种基于GPU的高效NDN数据名查找算法
评论
全部评论0/1000