Research On Performance Optimization Of Edge Computing For Data Intensive AI Computing

Edge computing has become a new computing service mode with its low latency and high performance.It deploys the resources of the cloud service center in the vicinity of the terminal device,and users can unload tasks to the edge server for execution,thus reducing the computing pressure of the cloud center,and on the other hand,reducing the transmission delay of tasks.Due to the heterogeneity and limitation of edge cluster resources,as well as the increasing data size and computational complexity,the traditional edge computing task scheduling methods no longer meet the user’s needs for low latency and low energy consumption.Therefore,this paper studies the optimization of computing performance of data-intensive AI tasks in the edge environment.In order to improve the performance of the edge system,the following three aspects are mainly studied:(1)In order to optimize the computing performance of data-intensive AI tasks in edge environments,this paper proposes a particle swarm optimization task scheduling algorithm based on target sequencing(TS-MOPSO).Firstly,a data-intensive AI task scheduling model is constructed,and the task scheduling is realized by combining multi-objective technology and particle swarm optimization algorithm.The algorithm introduces adaptive inertia weight and shrinkage factor update mechanism to improve the convergence speed of the algorithm.In addition,a new method based on target ranking is proposed,which can comprehensively weigh priority and conflicting optimization objectives,thus simplifying the search for the best scheduling scheme.Finally,the relevant attributes of data-intensive AI tasks are extended through the Cloudsim simulation platform,and the algorithms implemented by programming are embedded into the simulation framework to achieve task scheduling.Compared with the other three benchmark algorithms,TS-MOPSO can reduce the task delay and energy consumption,and improve the execution efficiency of the system.(2)In order to improve the computing performance of GPU-CPU heterogeneous computing system in edge environment,this paper proposes a genetic task scheduling algorithm based on reinforcement learning(QL-GA).First,the distribution of tasks on CPU and GPU processors under load balancing is predicted.QL-GA algorithm improves the disadvantages of early convergence and local optimization of genetic algorithm,and uses Q-learning to realize adaptive adjustment of parameters.Extend Cloudsim to GPUCloudsim module to realize a simulation architecture that supports GPU-CPU heterogeneous computing task scheduling.Then we evaluate QL-GA based on this architecture to verify the effectiveness of its algorithm in heterogeneous systems.(3)In order to improve the utilization of edge cluster resources,this paper uses virtualization container technology to effectively manage the resources in the edge environment.Firstly,a container deployment model based on computing job resource prediction is proposed to predict the computing mode and resource requirements of tasks in advance.On this basis,a dynamic deployment algorithm of container resources is proposed to adjust the computing shape and size of the container in real time.Extending the Container Cloud Sim module in the Cloudsim simulation platform and verifying the performance of the algorithm,the deployment model can realize the function of dynamic customization of different task execution environments,thus improving the resource utilization of the system.