Design And Implementation Of Load Balancing System Based On Microservices And Containers

In recent years,with the rapid development of Internet technology and the corresponding increase in network users,the traditional system is facing enormous pressure.The traditional system adopts a fixed resource model,with applications being deployed on physical machines,which are usually overused to reach the rated workload,resulting in low system availability and stability.It turns out that this model wastes energy and space,resulting in low resource utilization.Therefore,it is of great significance to study the load balancing system to make the entire application system highly available,stable with a high-resource utilization.At present,container technology and microservice architecture have been widely studied and applied in current network systems.Internet technology continues to iterate:one aspect is from the initial virtual machine deployment to today's more fine-grained container deployment,which have replaced virtual michaines due to its advantages of light weight,convenience and high resource utilization,another aspect of the iteration is from the initial monolithic architecture to microservices Architecture,with each service in the microservice architecture having its own independent function,and maintains a larger system together,owning a clear division of work.Consequently,it has significant advantages in development,testing,deployment and scheduling.Considering the two aspects abovementioned,this thesis aims to combines the above advantages to design and implement a load balancing system based on the microservice Spring Cloud framework and Docker container technology.The main work of the thesis is as follows:(1)A load-based bidirectional long-short-term memory network(Attention-Bi LSTM)prediction model with attention mechanism is proposed.Since traditional time prediction algorithms mostly use regression fitting with low accuracy,this thesis adopts the Bi LSTM neural network model and adds an attention mechanism to extract and care about important data features of time series,so as to improve the accuracy of prediction and the time efficiency of iteration.In addition the load model fully considers factors such as CPU,IO,memory and network,and the representation is stronger.The experimental results show that the model has good prediction accuracy and computational efficiency.(2)An improved discrete cuckoo optimization algorithm(RDCS)based on container dependence is proposed.When the system load is not balanced,container scheduling is required.The thesis deeply studies the dependencies between containers under the microservice architecture,improves the traditional Cucoo Search algorithm based on container dependencies,and adds roulette and container dependencies in the key steps of Levy flying and abandoning the new bird's nest,which enhances search power and efficiency of the global and local algorithms Finally,after verification,the algorithm improves the resource utilization and response speed of the system.(3)Design the load balancing system and apply it to the wind farm management system.Demand analysis for the load balancing system has the following functions:service high availability of service,resource monitoring,cluster service management and load balancing modules.In terms of architecture design,the modular divide-and-conquer structure is adopted to reduce the coupling degree of the system;the improved load prediction model and container scheduling algorithm are applied to the actual wind farm management system,which not only can provide stable external services,but also has the function of load balancing.After experimental verification,the load balancing system makes the application system respond faster and more stable.