Microservice Scheduling Strategy For Computing-aware Networks

With the rapid development of cloud computing and edge computing,the computing power has been extended from the cloud to the edge,forming a ubiquitous computing resource.In order to promote the in-depth integration of computing and network resources,the concept of computing-aware networks was proposed by researchers.In the architecture of the computing-aware networks,the service layer is a bridge connecting computing resources and users' requirements.It can be implemented based on microservice architecture,providing specific service instances for users' requirements flexibly.Therefore,it is of great significance to research on the scheduling strategies of microservice for the further implementation of computing-aware networks.Currently,in cloud center,the container orchestration of Kubernetes and Service Mesh represented by Istio are the mainstream platforms for microservice deployment and scheduling.However,in the aspect of microservice deployment,the information of CPU and memory is the only evaluation index of Kubernetes default strategy,so the other dimensional resource information such as the statuses of network and disk are not considered,resulting in unbalanced resource load.In the aspect of microservice scheduling,when a large number of requirements arrive,the default algorithm of Round Robin tends to degrade scheduling performance in Istio,due to the lack of efficient algorithm.Based on the service layer of computing-aware networks,optimization of the deployment and scheduling of microservice in cloud center is studied.Then,the optimization strategy proposed is migrated to the edge to synergize each other.The detailed work of this thesis is as follows:(1)An optimization strategy based on multidimensional resources for the microservice deployment is designed.When microservices are deployed by Kubernetes,due to the insufficient utilization of resource information in defult strategy,the optimization strategy is proposed.The strategy expands the evaluation index of network bandwidth and disk capacity information,and can dynamically adjust weights of multidimensional resources according to the different requirements.Also,a real-time monitoring of cluster resource status information is realized to make the decision more accruately.The experimental results show that compared with the defult strategy,the deployment optimization strategy proposed in this thesis can help to achieve a more balanced utilization of the resources for the nodes and the cluster.(2)A microservice scheduling optimization strategy based on critical path is proposed.To solve the inefficiency of service scheduling caused by use of the default algorithm of Round Robin,the scheduling problem is abstracted,and an AOV topology graph is constructed.Then the AOV graph is converted to an AOE model accroding to the processing time and cost of microservice.Meanwhile,a solution based on critical path method is proposed,that is,decides the lowest-cost service instance by considering the completion time of service instance on critical path nodes as the deadline.The simulation results show that the scheduling optimization strategy proposed can effectively reduce the time and cost of service instance scheduling compared with Istio default algorithm.(3)A scheduling optimization strategy of microservice is extended to edge cluster.Currently,for the problems of multi-node information maintenance and lack of fine-grained service scheduling capability on edges,a scheme of centralized edge information maintenance is designed in this thesis.Also,a lightweight Service Mesh for edge is adapted to achieve finegrained service scheduling on edge clusters.Finally,according to the differences between edge and cloud,the scheduling strategy proposed is adjusted to fit the edge.The results show that,on the edge side,the optimization strategy given in this thesis can still effectively achieve edge synergy for mircoservice scheduling.In summary,the optimization of the deployment and scheduling of microservices for computing-aware networks is implemented in this thesis.The proposed deployment optimization strategy can actively adapt to the changes of computing resources and balance the utilization of resources.The proposed scheduling optimization strategy based on critical path can significantly reduce scheduling time and cost,and has good performance on the edge side.Based on the above optimization strategies,the synergistic processing capability of microservices in computing-aware networks is effectively improved.