Autonomous Scheduling Strategies In Cloud And Edge Network

With the development of cloud computing and related techniques,more and more data centers in large scale are built.Cloud providers abstract these computation resources,encapsulate them into virtual machines in various granularities and lease them in a pay-as-you-go manner.Meanwhile,more and more enterprises,even individuals,start to rent these computation resource,deploy their applications onto it and provide services to their users.When they participate into cloud computing in role of cloud service providers(CSP)and establish their business upon the cloud,they can rescue themselves from the technique details and the mainte-nance of hardwares.However,it also put forward much higher requirements for the framework for cloud computing.Generally,cloud services constrain their quality of service(QoS)in the form of Service Level Agreements(SLA)from the perspectives of performance,availability,security,etc.This means that the framework for cloud computing needs to provide flexibility for deployed applications based on the SLA to achieve higher resource utilization,cost savings,and green computing.At the same time,it must maintain sufficient transparency and avoid unnecessary human intervention by scheduling according to some autonomous strategies,so that the user experience is improved as much as possible.Such scheduling mainly includes three processes,namely,resource provision,resource placement and resource dynamic consolidation.On the basis of the results and accumulations of my previous work,this paper complements the existing shortcomings and defects,and obtains a flexible framework for cloud computing,which completely includes the above three processes.Furthermore,it fully utilized an auto-regressive restriction model and a two-tier supply and demand relationship so that the pre-modeling of the applications,which aims at building the mapping from workloads to the amount of resources required,can be avoided.Meanwhile,with the development of the Internet of Things(IoT),the boundaries between data producers and data consumers have been broken.More and more new devices connected into the network now are equipped with computation capabilities.In order to save the bandwidth of the backbone network and improve user experience,including but not limited to reducing service delay,improving service security,and ensuring service privacy,edge computing,as an extension and expansion of cloud computing,is envisioned as a promising enabler to leverage computation capabilities at the edge.It is drawing more and more attentions.Computation offloading problem is one of the key problems in the research of edge computing.This paper proposed an advanced decision-making model which enables two-way initiative offloading in edge computing.This decision-making model utilizes the intrinsic topology at the edge network and entitles both sender and receiver of a given offloading operation with the ability of being conscious of the utilization of itself and the remote node,so that the model can effectively tolerate the occurrence of congestion and the expiration and failure of information,and make the decision-making more proactive and reasonable.The constraint of SLA of the cloud services becomes the bridge between the central cloud and the edge network.After analysis of the subtle but key differences between the central cloud and the edge network,this paper further makes the offloading process aware of the SLA and deadlines of different tasks,so that the central cloud can fully utilize the computation capabilities at the edge network,and the edge network can also effectively take advantages of the elasticity of the central cloud.These two thus can work together and be organically unified.Finally,the occupation of resources at the central cloud in runtime is reduced as much as possible and the overall resource utilization of the whole network is improved,while the transparency and the quality of the services are constrained.This paper carried out step-by-step evaluations on these above works,including:a com-parison evaluation on the aforementioned enhancements of the framework for cloud computing;detailed evaluation on the advanced decision-making model at the edge network;and an overall performance evaluation on the central cloud and the edge network.The results of these evaluations are inline with the expectations and the original intentions of the design.The correctness and effectiveness of these works are verified.