基于动态系数的三次指数平滑算法负载预测

The load prediction of three exponential smoothing algorithm based on dynamic coefficients.

摘要:

数据中心是企业信息化的重要组成部分,云计算的核心思想就是把数据中心整成一个资源池,对资源池进行统一的调度与管理。随着虚拟化技术的发展,目前对数据中心的资源利用率越来越高,但是还是存在大量资源浪费的情况,其原因在于当前对数据中心未来负载预测的算法还存在一定的局限性,如果对未来负载预测值远远大于实际负载情况,则导致大量的虚拟机资源利用率不高,反之则会导致虚拟机的资源使用率消耗增大,云平台中不同物理服务器之间的负载情况不平衡,一部分物理服务器负载过大,导致云计算平台响应时间过长。因此云计算平台选取一个合适的负载预测算法显得越发重要,如何权衡以上问题,是云计算里面的一个重点研究方向。负载预测选取时间序列预测算法中的三次指数平滑法,在该算法原有的静态系数基础之上,设计了一种动态系数提取方法。通过等距法把静态系数分成若干份进行训练,然后在预测过程中提取该时段误差最小值所对应的系数。在预测结束后,重新计算其误差,并通过均值法覆盖旧误差。实验结果表明,基于自适应三次指数平滑算法其预测误差明显小于静态系数所预测的误差,计算复杂度低,具有一定的应用价值。

Abstract:

data center is an important part of enterprise informatization, and the core idea of cloud computing is to integrate data center into a resource pool, and to conduct unified scheduling and management of resource pools. With the development of virtualization technology, the current resource utilization of the data center is higher and higher, but there are still a lot of waste of resources situation, the reason for this is that the current load forecasting algorithm for data center in the future also has certain limitation, if the load forecast for the future is far greater than the actual load, leading to a large number of the virtual machine resources utilization rate is not high, the opposite can lead to increase consumption of resource utilization, the virtual machine between different physical servers in the cloud platform of load imbalance, part of the physical server load is too large, lead to cloud computing platform response time is too long. Therefore, it is more and more important to select a suitable load prediction algorithm for cloud computing platform. How to balance the above problems is a key research direction in cloud computing. Based on the static coefficients of the algorithm, a dynamic coefficient extraction method is designed. By equidistant method, the static coefficients are divided into several parts for training, and then the corresponding coefficients of the minimum error of the time period are extracted in the prediction process. After the prediction is over, the error is recalculated and the old error is covered by means of means. The experimental results show that the prediction error is significantly less than the prediction error of the static coefficient based on the adaptive cubic index smoothing algorithm, and the computational complexity is low, which has certain application value.