Elastic Container Provisioning Methods In Fog Computing

Internet of Things(Io T)applications such as Smart Agriculture,Smart Factories and Smart Cities are becoming more and more widespread.However,due to the wide geographical distribution of Io T applications,even large-scale Cloud Computing platforms with strong computing power and large storage capacity cannot overcome the service delay problem caused by distance.Fog Computing(FC)extends Clouds to the edge of the network closer to users,including edge routers,network base stations and small data centers.Using the computing power of fog nodes can reduce latency and improve the Quality of Service(QoS).At the same time,it is also necessary to use the scalability of Cloud services to control the rental cost in order to give full play to the advantages of the FC.In the FC environment,Web applications usually provide business processing capabilities by a set of micro-services that cooperate with each other.Designing an elastic resource provisioning algorithm for meshed Web applications with complex calling relationships in FC is a hotspot of current research.The main difficulty is how to coordinate and scale geographically distributed resources of multiple data centers.As containers are lightweight and fine-grained,they are more suitable for meshed microservice systems.Kubernetes,a container orchestration platform,can be used to automate the deployment of containerized applications.To provide high-quality services for users from different regions,this thesis adopts a distributed and collaborative resource provisioning strategy to flexibly allocate an appropriate number of containers to multiple data centers.The main work of this thesis concludes: i)Proposing a container auto-scale method based on an adaptive processing rate queuing network;For a Kubernetes-based Cloud meshed microservice system,consider the negative impact of synchronous calls on container performance;Adopt queuing-length aware Jackson queuing networks,which quantifies the impact of the bottleneck tier on other micro-services;ii)Proposing a control method to allocate elastic container resources based on the Deep Neural Networks(DNN);Considering that the response time of requests is associate with business types,the accessing path based log analysis method is used to obtain the processing time of the service itself;Using DNN to establish a nonlinear model of the number of containers,request arrival rate and mean response time.For the meshed micro-service system in Clouds,a container auto-scale method based on DNN performance model,queuing model and control theory is proposed;iii)Proposing a resource provisioning method for meshed micro-service systems in FC environment that considers the delay between fog nodes in different regions;Considering the delay between different nodes when calculating the actual average service time of services to guarantee SLA(Service Level Agreement)from different regions;(4)Building a real test platform based on Kubernetes and a resource control module has been developed to provide a testbed for different resource provisioning algorithms;Using real load data and a meshed service system based on Kubernetes,this topic has compared the proposed method of container resource supply in FC environment with existing algorithms.Experiments show that the method proposed in this thesis can effectively reduce the rental cost and guarantee the QoS.