刘开华，闫 柳，李 卓，宫霄霖，彭 鹏，王彬志
AuthorsHTML:刘开华，闫 柳，李 卓，宫霄霖，彭 鹏，王彬志
AuthorsListE:Liu Kaihua，Yan Liu，Li Zhuo，Gong Xiaolin，Peng Peng，Wang Binzhi
AuthorsHTMLE:Liu Kaihua，Yan Liu，Li Zhuo，Gong Xiaolin，Peng Peng，Wang Binzhi
Unit:天津大学微电子学院，天津 300072
Unit_EngLish:School of Microelectronics，Tianjin University，Tianjin 300072，China
Abstract_Chinese:针对命名数据网转发信息库快速检索差异化名称数据、高效存储转发信息和有效支持最长名称前缀匹配机制的需求和挑战，提出了基于神经网络的命名数据网学习型FIB整体方案，称L-FIB．首先，介绍了L-FIB的索引结构Learning Tree，通过使用塔式两级神经网络模型学习索引内容在存储器中的分布情况，实现更均匀的数据映射，降低映射冲突，提高存储效率．其次，研究了L-FIB的存储结构和名称数据检索算法，片内高速存储器部署多个与不同名称前缀组件数相对应的索引结构Learning Tree，片外低速存储器部署多个与索引结构Learning Tree对应的FIB存储池，并通过相应的名称数据检索算法实现对兴趣包的转发信息检索和转发信息更新操作，有效支持了命名数据网的最长名称前缀匹配机制，提高了名称数据检索速度．实验结果表明，L-FIB在误判概率、存储消耗和吞吐量方面的综合性能明显优于其他对比方案．在误判概率低于1%的条件下，L-FIB的索引结构存储消耗仅为58.258MB，能够部署于高速存储器SRAM上．L-FIB的实际吞吐量约为11.64×10⁶数据包/s，可以满足当前命名数据网对数据包快速处理的要求．
Abstract_English:Designing an effective forwarding information base(FIB)for named data networking(NDN)is a major challenge within the overall NDN research area，since an FIB has to perform fast lookups for complex names，provide high capacity，and accurately support the mechanism of longest name prefix matching (LNPM). Therefore，a learning FIB based on neural networks，called L-FIB，was proposed. First，the index of L-FIB，named Learning Tree，used a two-level neural network model to learn the distribution characteristic of data indexed in static memory，which achieved more uniform mapping，reduced the false positive probability and improved memory utili-zation. Second，the storage structure and name lookup algorithms of L-FIB were put forward. The on-chip memory using SRAMs deployed multiple Learning Trees corresponding to the name prefixes with different numbers of compo-nents，while the off-chip memory using DRAMs deployed multiple FIB stores corresponding to the Learning Trees. The name lookup algorithms were also described to implement the retrieval of forwarding information for the Interest packets and the update of forwarding information. This well supported the LNPM mechanism and realized fast name lookups. Experimental results showed that the overall performance of L-FIB was superior to the compared schemes in terms of false positive probability，memory consumption，and the throughput. The index of L-FIB significantly re-duced memory consumption to 58.258MB with the probability of false positive＜1%，which meant it was deployable on SRAMs in commercial line cards. The throughput of L-FIB was about 11.64 million packets per second，which met current network requirements for fast packet processing.
Keyword_Chinese:命名数据网；转发信息库；神经网络；名称数据检索
Keywords_English:named data networking(NDN)；forwarding information base(FIB)；neural network；name lookup

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6498