| The arrival of 5G era has promoted the development of CAAS platform architecture based on "container technology",but the solution to the problems of high energy consumption and low resource utilization of cloud platform is becoming more and more urgent.In addition,with the introduction of micro-service architecture and service choreography,the service of cloud platform has become more complex and the probability of resource failure has increased significantly.Therefore,fault tolerance during deployment and scheduling becomes an inevitable problem for application systems in the process of "going to the cloud".In view of the above problems,this paper mainly focuses on the resource scheduling strategy,including the following three aspects.First,in the study of fault-tolerant static container placement,this thesis proposes a container selection strategy based on dominant resource balance to improve resource utilization and reduce the waste of fault-tolerant resources.This strategy integrates 2-DRF idea into the discrete degree formula as the container scoring strategy,which not only ensures the balance of resource utilization,but also improves the utilization rate.In addition,a container replica-first placement strategy is proposed to solve the problem of under-utilization of host resources caused by fault-tolerant constraints in container deployment.Finally,experimental results show that this strategy not only meets the requirements of fault tolerance,but also reduces the host occupancy by 5.7%,and is superior to other algorithms in terms of energy consumption,resource utilization and SLA default rate.Second,in the study of dynamic container migration based on fault tolerance,this thesis defines the resource utilization of cloud platform as different states,and formulates different optimization strategies for different states.At the same time,a strategy of full container resource migration based on dynamic threshold is developed for the"fragmentalized" cloud platform.In terms of out-migration,a soft threshold container out-migration strategy based on the proportion correlation of dominant resources is proposed.In the aspect of migration,the container placement algorithm based on dominant resource equity is adopted.In addition,a migration timing strategy based on resource fragmentation evaluation algorithm is proposed to reduce the number of container migration.Experiments show that this algorithm is more suitable for cloud environment with large fluctuation of resource demand,and it reduces the energy consumption by 5%on average compared with other algorithms.Thirdly,in the study of virtual machine migration scheduling,this thesis proposes a hybrid virtual machine migration strategy which combines the dominant resource optimization idea and the improved discrete particle swarm optimization algorithm.In terms of multi-objective optimization,this strategy combines the ideas of Pareto optimization and 2-DRF.In terms of improving particle swarm defects,this strategy absorbs the advantages of SA and NSGA-II algorithms,and increases diversity and domain search.In addition,the strategy also combines the discrete characteristics of DPSO,and on this basis,the position,velocity and update formula of particles are redefined.Experimental results show that the proposed strategy can reduce the number of hosts occupied and the energy consumption of cluster data center by 12.5%,which is better than similar algorithms.To sum up,this thesis mainly studies the resource scheduling optimization of container cloud platform,in which the scheduling resources include containers and virtual machines.The resource scheduling of the container is based on the virtual machine,which together affects the availability and low energy consumption of cloud services.It can be seen from the experimental results that the proposed algorithm meets the user's requirements for fault-tolerant rules,reduces the host occupancy at run time,and further reduces the overall energy consumption of the cluster. |
PDF Full Text Request