Service Quality Monitoring And Decision Making Of Elastic Scaling In Cloud

With the fast development of computer science and technology, cloudcomputing becomes one of the hottest topic in computer science. The reasonof the wide concern and application of cloud computing is that cloudplatforms provide the ability of elasticity and elasticity is one of the mostimportant standards to evaluate a cloud platform. By elasticity, we mean thatcloud platforms support the dynamic scaling of resources. In order to makecloud platforms own the ability of elasticity, we need to monitor the servicesand make decision of the elastic scaling according to the monitoring result.So the service quality monitoring and decision making of elastic scaling isvery important to cloud platforms.For the service quality monitoring problem, the monitored systems arebasically distributed systems. The distribution of metrics causes theproduction of communication overhead when obtaining the overall servicemetrics. And the continuous variation of metrics disables us from gettingreal-time metrics continuously and remotely. For the elastic scaling decisionmaking problem, because resource scaling in cloud platforms needs a lagtime which cannot be ignored, we need to predict future metrics in some wayand make the decision of elastic scaling ahead of time in order to ensureservice quality during the scaling process. So the goal of this paper is topropose a service quality monitoring and elastic scaling decision makingsolution which is suitable for cloud platforms. This solution should monitorservice quality in a communication efficient way and make elastic scalingdecision promptly according to the monitoring result and under theconsideration of the scaling lag time.Aiming at above problems, we propose our approach which includes two monitoring and decision making strategies in this paper, which are thebasic strategy and the slack scaling out trigger line strategy. In the basicstrategy, we monitor the increasing speed of resource usage in acommunication efficient way. And we propose the concept of “scaling outtrigger line”. We make decision of elastic scaling according to monitoringresult and the scaling out trigger line. Then, based on the basic strategy, wefurther propose the slack scaling out trigger line strategy. We control theupdating of the scaling out trigger line and further reduce the communicationoverhead. Besides, the resource scaling size problem and the scaling inproblem are discussed briefly in this paper.Based on the above approach, we design and implement the servicequality monitoring and elastic scaling decision making solution. The designand implementation of this solution has good modifiability and extendibilitydue to the adoption of the “high cohesion and low coupling” thought.Afterwards, experiments are performed to evaluate our approach and thedesign and implementation. Several experiments are performed in aspects ofcommunication overhead and scaling decision making under differentworkload patterns and parameters. The result shows that our approachreduces the communication overhead efficiently and can make decision ofelastic scaling more promptly than trivial method.