基于改进型MBF的命名数据网PIT存储结构研究

摘要

Named data networking (NDN) is a future network architecture to solve the problems faced by TCP/IP(transmission control protocol/internet protocol).In forwarding plane of NDN router,a novel structure named pending interest table (PIT) is employed to store the forwarding information of pending interest and forward data with the information.Therefore,designing an efficient PIT is important for improving the forwarding performance.In this paper,an improved data structure called bitmap-mapping bloom filter (B-MBF) is proposed,which can obtain multiple hash values by one hash function thus boosting the lookup speed,and dynamically allocate the offset address for each element based on Bitmap.Meanwhile,a PIT structure named B-MaPIT is presented and the memory consumption can be changed with the number of elements by managing multiple dynamic memory spaces.The experimental results show that B-MaPIT has better performance compared with the existing PIT structures in terms of the memory consumption,building speed and throughput.%命名数据网(named data networking,NDN)是一种新型的未来网络架构,以解决当今TCP/IP(transmissioncontrol protocol/internet protocol)网络面临的问题.NDN路由器的转发平面采用了一个新的数据结构待定Interest表(pending interest table,PIT),用于记录所有待定兴趣包的转发信息,并根据此信息正确地转发返回的数据包.因此,设计一个高效的PIT存储结构对提高转发性能具有重要的作用.结合Bitmap提出一种改进的数据结构B-MBF(bitmap-mapping bloom filter),该结构采用一个哈希函数实现多次哈希映射以提高检索速度,并利用Bitmap实现元素内存单元的地址偏移量的动态分配.同时,在此基础上提出PIT存储结构B-MaPIT,通过管理多个动态存储空间实现片外存储消耗随元素数量改变而改变.通过仿真实验与现有PIT结构进行了性能对比,结果表明B-MaPIT在存储消耗、数据表构建速度、吞吐量方面具有更优秀的性能表现.