Elasticity Evaluation In Cloud Computing Based On Pattern Matching Analysis

At present,cloud computing has emerged and gradually becomes a mainstream computing paradigm,the companies which use cloud platform as the main IT infrastructure are gradually increasing,these trends make the core value of cloud computing changed from the cost reduction to the agility and liberalization,while achieving agility and liberalization requires elastic resource management strategies.Elasticity has become the core function of resource management in cloud service,and elasticity evaluation has become a hot issue in this field.However,the cloud platform and the traditional IT computing platform are showing a significant difference in both architecture and development process,the original benchmarking method can not meet the needs of elasticity evaluation.But at this stage a comprehensive and reasonable elasticity evaluation system has not yet been fully constructed,the main difficulties are from two aspects: firstly,there are a number of elasticity evaluation indicators,choosing a smaller and containing the core indicators of the evaluation index set is the main way to solve such problems;Secondly,the evaluation index has to be refined,there exists the problems that the system state is not accurate and the delay time is processed fuzzily,which affects the accuracy of elastic evaluation.In order to solve the problem of quantification and comparison of elasticity in cloud environment,and improve the insufficiencies in system scheduling delay and state definition.In this paper,we analyze the core features of elasticity,and put forward new metrics that capture the precision of resource provisioning and de-provisioning,as well as the timing aspects of an auto-scaling mechanism,explicitly.On this basis,we establish a quantitative model from the aspects of system availability,the speed and precision of scaling to describe the elasticity more accurately.For the purpose of refining these metrics,we use the change point detection(CPD)technique and the dynamic time warping(DTW)to carry out the time series data mining.Through these pattern-match analyses,we tackle the problem that the data provided by cloud systems for users is incomplete and inaccurate,providing support for the precise calculation of the metrics.To validate the proposal,we sketched a new elasticity tailored benchmarking methodology to quantify the degree of elasticity,and evaluate the most representative strategies of the two mainstream scheduling forms which are reactive and proactive.Finally,we demonstrate that the proposed approach is feasible and applicable in different cloud scenarios.