Abstract: For the lower efficiency of resource allocation and lacking precise prediction for resource requirement, this paper proposes a resource allocation approach based on mixed prediction. According to the cyclical feature of resource requirement, this proposed approach employs FFT (fast Fourier transform) tools to determine the cyclical attribution. If there is no such attribution existed, the Markov chain is alternatively used to predict the tendency of resource requirement. The experimental results show that the proposed approach can predict the future resource requirement more precisely. Moreover, the proposed approach can also allocate the virtual machine resource adaptively, decrease the number of occupied physical machines, and reduce the number of violating the SLA (service-level agreement).

Key words: virtual machine, resource allocation, fast Fourier transform, Markov chain

摘要： 针对云平台下资源利用率低，缺乏资源变化的准确预测等问题，提出了基于混合预测的云平台资源分配方法。该方法根据服务资源需求的周期性特点，采用快速傅里叶变换（fast Fourier transform，FFT）的手段判断服务资源需求的周期性，对缺乏周期性的资源请求采用马尔科夫过程进行预测，进而可依据预测结果自适应地分配虚拟机资源。实验结果表明，该方法能够准确预测服务资源需求，合理分配虚拟机资源，提高了虚拟机资源利用率，有效降低了SLA（service-level agreement）的违反次数。

关键词: 虚拟机, 资源分配, 快速傅里叶变换, 马尔科夫链