Research On Optimization Of Microservice Load Balancing Algorithm Based On Hardware Metrics

In recent years,with the rapid development of microservice technology,the microservice architecture has been widely applied by many internet companies.However,as the demand for services increases,the problems of low resource utilization and high energy consumption in server clusters have become increasingly significant.Prolonged high-load status of servers can increase the risk of crashes,which can severely affect the stability of the system.Therefore,it is of great research significance to propose a reasonable load prediction scheme and load balancing algorithm.This article proposes a hardware-based microservice load balancing algorithm,which uses a load prediction neural network model to predict the load value of each server.Based on the predicted load value,the algorithm is optimized to improve the concurrency of microservice servers.The main research work is as follows.(1)This paper proposes using CRITIC to calculate the server load values.The objective weighting method is used to obtain the weights of hardware metrics such as CPU usage,memory usage,and disk IO,and the time series data of server load values are calculated.CRITIC is used to weight the various hardware metrics of the server,thereby obtaining the server load value and predicting the server load status in the next moment more effectively.(2)Built a load prediction model based on dual attention mechanism.This paper constructs a load prediction model based on the long short-term memory(LSTM)network with a dual-attention mechanism,which introduces feature attention and time-series attention mechanisms.The dual-attention mechanism enhances the network’s ability to capture temporal and feature information,effectively improving the accuracy of single-step and multiple-step predictions.Experiments are conducted using Alibaba’s Cluster-trace-v2018 public dataset,and the dual-attention mechanism network is compared with RNN,LSTM,GRU,and other prediction networks.The results show that the dual-attention mechanism network has better stability and accuracy.(3)Proposed a microservice load balancing algorithm based on hardware metrics.Traditional load balancing algorithms cannot reasonably allocate resources based on the real-time load status of servers,resulting in uneven resource utilization.This paper proposes a hardware-based microservice load balancing algorithm that monitors the CPU usage,memory usage,disk IO,and other hardware metrics of server clusters and predicts the load value of each server.By assigning different weights to the predicted load values for the next three time steps as the weight judgment conditions for the load balancing algorithm,user requests are reasonably allocated.Experimental results show that compared with other load balancing algorithms,the proposed algorithm has stronger anti-concurrency ability and stability.