Font Size: a A A

Research On Key Techniques Of Auto-scaling Docker Cluster Based On A Combined Prediction Model

Posted on:2019-10-15Degree:MasterType:Thesis
Country:ChinaCandidate:T Z WangFull Text:PDF
GTID:2428330572950774Subject:Computer technology
Abstract/Summary:PDF Full Text Request
With the rapid development of cloud computing,virtualization technology supporting cloud computing is also more mature.Docker container virtualization technology is currently a research hotspot in this field.Docker containers have the advantages of standardized application release,low computing resource consumption,rapid deployment and startup,and are one of the mainstream technologies for hosting cloud platforms.However,concurrent access to cloud services tends to be bursty.At this time,the workload of the cloud platform will increase drastically,and it is necessary to rapidly increase computing resources to maintain service quality.Similarly,when traffic decreases,it is also necessary to reduce idle resources according to the amount of load to reduce costs.Therefore,the automatic scaling function is very important for the cloud platform.The specific work of this article is as follows:(1)This article analyzes the key issues facing the Docker cluster construction,and solves the problem of network communication and mirror sharing of the Docker cluster,using Calico and Harbor respectively to solve,and realize the Docker cluster construction.(2)Based on the study of GM(1,1)model and BP neural network model,this paper proposes a combined forecasting model,which uses BP neural network model to improve the nonlinear error for the forecasting result of the grey forecasting model,and makes the forecasting result consider linearity.Two aspects of nonlinearity.And according to the prediction result of this combined prediction model,resource scheduling is performed.(3)For resource monitoring,load balancing and service discovery problems of clusters,a resource monitoring subsystem based on CAdvisor and Heapster,high availability load balancing based on Haproxy,and automatic service discovery based on Consul are established.Based on these modules,combined forecasting is used.The model implements a Docker cluster scaling based on the prediction results.(4)Through experiments,the prediction accuracy of the combined forecasting model of GM(1,1)and BP neural network is firstly verified,and it is proved that the combined forecasting model of this paper has better prediction accuracy.An auto-scaling Docker container cluster based on workload prediction was built,and the scalability of the clusterwas tested,demonstrating the effectiveness of cluster auto-scaling.
Keywords/Search Tags:Cloud Computing, Workload Prediction, Auto-scaling, Docker
PDF Full Text Request
Related items