面向智能路由的多级哈希网络数据存储结构

网络数据的采集和存储是智能路由控制的基础，为智能路由提供了大量的网络流量数据进行模型训练和决策。然而，作为网络数据存储系统中的核心设备，交换机的存储空间非常有限，且设计灵活性低，无法满足智能路由控制对全面高精度的数据存储和轻量级存储系统的需求，进而影响智能路由控制的效果。提出一种面向智能路由控制的多级哈希网络数据存储结构，高效利用交换机有限的存储空间，实现低碰撞率的网络数据存储。该结构通过多级哈希表增加数据的可存储空间数量，从而降低存储冲突率并提高存储空间利用率。同时，该结构使用基于低开销时间戳的LRU算法解决哈希冲突：在发生哈希冲突时总是保存最新的网络数据，清除陈旧数据，以尽可能减少后续的存储冲突。基于真实网络流量数据的实验证明了相比目前普遍使用的单级哈希存储结构，多级哈希存储结构在存储碰撞率和负载率两方面存在显著的性能优势。

Multi-stage hashing network storage structure for intelligent routing

Network data collection and storage are fundamental for intelligent routing control,providing massive flow data for model training and decision-making.However,as the key device in network storage system,switches have very limited memory size and design flexibility,which can’t satisfy the needs of intelligent routing control for comprehensive high-precision data and lightweight storage system,thus reducing the effectiveness of intelligent routing control.A multi-stage hashing network storage structure (MHNSS) was proposed for intelligent routing control,which fully utilized the limited memory resource of switch and completed network data storage with low collision rate.The number of candidate buckets of the flow key was augmented by multi-stage hash table,thus reducing collision rate and improving memory load ratio.Hash collision was resolved via coarse-grained timestamp LRU algorithm,which always stored most recently used data and cleared least recently used data to avoid subsequent collisions as far as possible.Trace-driven experiments showed that compared with widely used single hash table,MHNSS had significant performance advantage in collision rate and load ratio.