基于LSTM的标识解析节点负载均衡算法

A LSTM-based Load Balancing Algorithm for Identity Resolution Nodes

工业互联网标识解析二级节点是面向特定行业提供标识服务的公共节点,随着二级节点在众多行业之间的快速落地,标识的注册量和解析量将达到海量级别,二级节点系统时刻承担着高并发的标识服务请求带来的压力,可以通过负载均衡机制解决上述问题.然而传统的静态负载均衡方案没有考虑系统的实时运行情况以及流量的突发性,因此本文面向标识解析二级节点提出了一种基于长短期记忆(Long Short-Term Memory,LSTM)网络的动态负载均衡方案,该方案借助LSTM算法对服务器请求连接数进行预测,并结合服务器的实时性能指标以及负载信息来选择最佳节点处理用户请求.仿真实验结果表明:相比于传统的负载均衡方案,本文提出的方案不仅有较高的响应成功率,还将集群处理并发请求的平均响应时间降低了50%左右,将集群的吞吐率提升了30%左右.

The second-level node of identity resolution in industrial internet is a public node that provides identity services for specific industries.With the rapid development of second-level nodes among many industries,the registration and resolution of identities will reach massive levels,and the second-level node system will always bear the pressure caused by high concurrent requests of the identity service.It's necessary to solve the above problems through a load balancing mechanism.However,the traditional static load balancing scheme does not consider the real-time operation situation and the traffic surge of the system.Therefore,this paper proposes a dynamic load balancing scheme based on Long Short-Term Memory(LSTM)for the second-level node system of identity resolution in Industrial Internet of Things.This scheme uses the LSTM algorithm to predict the number of server connection requests,and combines the server's real-time performance indicators and load information to select the best node to handle users'requests.The simulation experiments show that the proposed scheme not only has a higher response success rate,but also reduces the average response time for processing the concurrent requests of the cluster up to about 50%,and increases the throughput of the cluster up to about 30%,compared with the traditional load balancing scheme.