一种面向SLA的云计算环境下虚拟资源调度方法

摘要

Application providers can choose the numbers of virtual resources to be rented on demand in cloud computing environment.To rent a large amount of resources will be returned in the QoS (quality of service)guarantee,therefore obtains better earnings,however the application providers have to pay for the resources renting.In reverse,to rent a few resources will save the investment in resources renting, but the insufficient resources are hard to guarantee the QoS,and the breach of SLA (service level agreement)will lead to economic losses. Therefore,it becomes a problem to be urgently solved that to rent how much virtual resources can the resources be put into fullplay and reach the earnings maximisation for the application providers.In light of these problems,taking the prospects of application providers’earnings into consideration,and balancing the earning losses of SLA and the rental cost investment of the servers,we present a virtual machine resource scheduling method aiming at maximising the earnings of the application leasers.The method models the performance of target cloud applications with queue theory,and introduces SLA earning losses function to estimate the SLA earning losses in circumstances of renting certain virtual resources.At last it uses hill-climbing algorithm to dynamically adjust the amount of physical resources renting so as to hit the goal of earnings maximisation.Experimental results demonstrate that,compared with the common virtual machine resources scheduling methods at present (1 VMPerJob,1 VM4All and BinPacking),the proposed method saves the costs invested by the cloud application providers by 1 7.92%,22.85% and 1 0.74% respectively.%云计算环境下，应用提供者可以根据需要决定租用虚拟资源的数量。租用大量资源会得到QoS保障，从而获得较好的收益，但应用提供者需要为租用资源付费。同样，租用少量资源，会节省租用资源投入，但资源不足难以保障QoS，SLA（Service Level Agreement）违背会带来经济损失。因此，租用多少虚拟资源才能使得资源得到充分利用，达到应用提供者的收益最大化成为亟待解决的问题。针对此问题，从应用提供者收益角度考虑，兼顾SLA收益损失和服务器租用成本投入，提出虚拟机资源调度方法，旨在使得应用租用者收益最大化。该方法利用排队论建模目标云应用的性能，并引入SLA收益损失函数估算在租用一定虚拟资源情况下的SLA收益损失，最后，利用爬山算法动态调整物理资源租用数量以达到收益最大化的目标。实验结果表明，与当前常见的虚拟机资源调度方法1 VMPerJob、1 VM4All和BinPacking相比，所提出的方法使得云应用提供者的成本投入分别降低了17．92％、22．85％和10．74％。