Research On Construction Of Elastic Cluster Based On Docker And Resource Pre-scheduling Strategy

In realistic scenario,the server will often confront unpredictable load demand,especially activities like Tianmao double 11 renew the record every year.The myriads of trading orders prove huge quantity of users access.During this period of time,the amount of servers ought to be increased rapidly to deal with extra load and decreased when the number of access declines,so that the efficiency of servers will be improved and the cost will be saved.Therefore,achieving auto scaling of the number of servers in cluster shows its great importance.Auto scaling is one of the most vital functions in cloud computing,of which virtualization technology is the cornerstone.In other words,the degree of virtualization determines the architecture of cloud platform.Currently as the virtualization of operating system level,Docker container technology provide new thoughts as it is achieved from the process level.Comparing with most cloud platform based on virtualization technology,Docker can offer faster start and stop velocity and better resource utilization rate.This thesis studies using Docker as the underlying virtualization technology of cloud platform.Through monitoring the cluster's real-time performance and scheduling the cluster's resource reasonably,constructing elastic cluster to be face with unpredictable load demand,so that keeping the cloud platform running in a low cost,high efficiency and stable status.Besides,improving the cloud platform's resource utilization rate,responding to network peak effectively and offering service that has high throughout and low delay.The main research work of this thesis includes:(1)Solving two main problems during the process of constructing container cluster: the communication between cross hosts and file sharing.With respect to the former,this thesis proposes three solutions: OVS combines with GRE tunnel technology,Flannel as well as DR(Direct Routing).Then solving the file sharing problem by constructing Docker registry and Network Fie System(NFS)of multiple nodes combined with DRBD.(2)Achieving performance monitoring of Docker elastic cluster.Specifically,collecting,saving,grouping,aggregating and showing the data of Docker containers and nodes based on OpenTSDB and container technology,providing judging condition ofwhether to extend or shrink the number of cluster's containers.(3)Constructing elastic scaling group.Employing Haproxy as load balancing service,Consul as automatic service discovery tool,Confd as the tool that gather information from consul and then generating Haproxy configuring file.(4)Studying and realizing several kinds of resource scheduling strategies,including the strategy that based on resource utilization rate,hosts scheduling and load predicting.(5)Constructing elastic cluster and testing the whole system to verify its availability and feasibility.