Research On Container Scheduling Methods In Cloud Data Centers

As the COVID-19 pandemic has affected the world,cloud services such as online shopping,telecommuting and online education have become a new link connecting people’s lives.Cloud computing and big data technologies provide critical support for COVID-19 analysis and prediction,virus tracing,prevention and treatment,and resource allocation.With advances of virtualization and container technologies,workloads in cloud data centers have evolved from physical servers to virtual machines and containers in a fine-grained manner.Cloud containers simplify the process of building,deploying,and running applications.This provides an opportunity for large-scale applications to upgrade from single architecture to distributed architecture,and containerization has become a mainstream trend.Large-scale Internet applications running on data centers are typically instantiated as a set of containers.Assigning a container to its affinity machine can reduce communication and transport costs while assigning it to the anti-affinity machine may affect the proper operation of the container.Existing container scheduling methods cannot accommodate these two types of requirements.To improve the operation efficiency and reduce the operation and maintenance cost of data centers,this paper focuses on the container instance allocation in heterogeneous server cluster.The main work in this paper includes the following two items:(1)This paper proposes a global cost-aware container scheduling strategy(GCCS)solve the container instance allocation problem.The purpose is to minimize the total power consumption of the cluster globally and meet the affinity requirements of applications.The ratio of the number of containers per server selected by the application is studied and modeled as an integer linear programming problem(ILP).Then a heuristic search algorithm is proposed to repair the slack solution of the ILP into an optimal solution.In addition,the Bayesian optimizer is used to perform several automatic development and exploration processes for the selection of cost coefficients,and the optimal cost coefficients recommended by Bayesian optimizer are used in the experiments.Experimental results show that GCCS can significantly reduce the total power consumption of the cluster while maintaining a high affinity satisfaction ratio.(2)The container instance allocation problem in this paper is mapped to a noncooperative evolutionary game scenario.Each application is considered as a player in the game and must constantly adjust its strategy over the course of evolution to increase its total payoff.This payoff takes into account power consumption,affinity and anti-affinity requirements.Then,the replicator dynamic equation of the model was given,and the evolution process is simulated by the explicit Runge-Kutta integral method.Finally,a container scheduling algorithm based on the mixed strategy evolutionary game(MSEG)is proposed.Numerical simulation results show that MSEG converges to steady state effectively.Several subsequent simulation experiments show that compared with other benchmark algorithms,MSEG can significantly improve the benefits of participants and show advantages in balancing the load of clusters.