Research On Automatic Elastic Extension Of Virtual Service Node Based On Scenario Driven

With the rapid development of social applications represented by self-media,the amount of data processed by cloud data centers per unit time increases exponentially,and due to the diversification of data sources,the increase or the decrease of data volume is affected by random factors and drasticly fluctuated.In order to accurately and timely provide resources to meet the demands of users,when and how to start the elastic mechanism of the cloud data center has become an important research issue.Based on the burst traffic scenario,an elastic method of cloud data center is proposed and it includes threshold learning,trend prediction,value prediction and feedback correction.The threshold learning algorithm self-adaptively learns the resource performance indicator threshold of the service node according to the different resource configurations;the trend prediction algorithm extracts the burst traffic's trend information,and determines the elastic start time;the value prediction algorithm construts model from the burst traffic's random information and obtains the predicted value of the burst flow;the feedback correction algorithm calculates the amount of resource expansion required for the elasticity.The trend prediction algorithm presented in this thesis improves the linear regression model,fits the trend factor of the burst flow,and combines the threshold obtained above to give the judgment of the elastic start time.The feedback correction algorithm proposed in this thesis calculates the feedback factor based on the resource usage of the current timeof the service node,and realizes the dynamic mapping from the burst traffic prediction value to the resource expansion amount.From the experimental results,the prediction algorithm proposed in this thesis is more accurate than the conventional methods such as LR algorithm,ARIMA algorithm and GARCH algorithm.The proposed threshold learning algorithm can accurately and autonomously learn resources threshold according to the different configurations of service nodes,it is more scientific and universal compared with manually setting thresholds.The proposed feedback correction algorithm improves the ability of guarantee the availability of business systems compared with conventional methods such as single threshold mechanism.