Research On Benchmark Of Elasticity Of Cloud Computing Platform

Cloud computing is a kind of services using and delivery models based onthe Internet. Cloud services promise to provide users with on-demand resources.Resources should have the ability of elastic expansion. Capabilities can beflexible configured, increased and released. Customers face unlimited servicecapability, and they can submit a certain degree of extension demands whennecessary. Elasticity is the key feature of cloud computing technology. Forcloud service with weak flexibility, the idle resources caused by over-provisionwill greatly increase the cost of both the service providers and customers, andthe lack of service capabilities caused by under-provision will affect the qualityof service and even lead to the failure of the service. The ideal provision is tomeet the SLA (Service Level Agreement) requirements with the smallestpossible amount of resources. Over-provision and under-provision are both therepresentation of weak elasticity. Elasticity is becoming an important indicatorof cloud computing.Benchmark is an evaluation through a series of standard tests and trials toassess the relative performance of the SUT (System Under Test). A benchmarkis composed of three parts: workloads, calculation models, and performanceindicators. Benchmarks provide a full range of comparison between the differentsolutions for customers and vendors to help them to choose the suitable solution.Some related test methodologies and suites for cloud computing have arisen.However, there have been no appropriate benchmarking methodologies andtools for elasticity especially dynamic elasticity.This paper first analyzed and gave the definition of elasticity. Then afterlooking into the lack of existing benchmarking methodologies in cloudcomputing environment, we designed the SLA-oriented and workload-driven elasticity benchmarking methodology and tool. We focused on the elasticitybenchmarking models and metrics, presented an elasticity benchmark design,including the design of the workloads, the calculation model as well as metrics.Based on the elasticity benchmark model above, we made further research tosolve a series of specific issues including elastic application design, run-timedata collection and test results analysis. Ultimately we implemented thebenchmark tool ADCBench. Finally, experiments for the elasticity evaluation ofthe laboratory’s cloud platform are carried out using ADCBench. The resultsshowed that the benchmarking methodology and tools achieved the anticipativeresearch target.