面向多模态数据的混合型FIB

针对未来网络转发信息库（FIB）中多模态数据带来的差异化快速索引、高效存储转发信息和多模态数据最长前缀匹配等问题，设计了一种支持多模态数据索引的混合型FIB，称为Hybrid-FIB.通过对不同类型的数据进行差异化处理，得到可供神经网络模型学习的输入向量，进而训练出能够实现均匀分布的神经网络混合索引模型.为了实现多模态数据最长前缀匹配，在片上静态随机存取存储器中部署两组Hybrid-FIB结构.实验结果表明，该混合型FIB在误判率、存储消耗及吞吐量等方面具备优异性能.

A Hybrid Forwarding Information Base for Multi-Modal Data

In order to solve the problems of rapid indexing, efficient storage of forwarding information and longest prefix matching brought by multi-modal data in the forwarding information base(FIB) in the future network, a hybrid FIB based on neural networks, called Hybrid-FIB, which supports multi-modal data indexing is designed. Hybrid-FIB differentiates different type of data to obtain input vectors for neural network model, and then trains a neural network hybrid index model that can achieve uniform distribution. Experiments show that deploying two sets of Hybrid-FIB on the static random access memory can not only achieve the longest prefix matching of the multi-modal data, but also have better retrieval speed and misjudgment rate than the current network.