Title: Load forecasting for cloud computing based on wavelet support vector machine

Authors: Wei Zhong; Yi Zhuang; Jian Sun; Jingjing Gu

Addresses: College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, 29 Jiangjun Rd., Nanjing, Jiangsu Province 211106, China ' College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, 29 Jiangjun Rd., Nanjing, Jiangsu Province 211106, China ' College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, 29 Jiangjun Rd., Nanjing, Jiangsu Province 211106, China ' College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, 29 Jiangjun Rd., Nanjing, Jiangsu Province 211106, China

Abstract: Because the tasks submitted by users have random and nonlinear characteristics in the cloud computing environment, it is very difficult to forecast the load in the cloud data centre. In this paper, we combine the wavelet transform and support vector machine (SVM) to propose a wavelet support vector machine load forecast (WSVMLF) model for the cloud computing. The model uses the wavelet transform to analyse the cycle and frequency of the input data while combining with the characteristics of the nonlinear regression of the SVM, so that the task load can be modelled more accurately. Then a WSVMLF algorithm is proposed, which can improve the accuracy of the cloud load prediction. Finally, the Google cloud computing centre data set was selected to test the WSVMLF model we proposed. The comparative experimental results show that the algorithm we proposed has a better performance and accuracy than the similar forecasting algorithms.

Keywords: nonlinear regression; task load; Google cloud computing; high performance computing.

DOI: 10.1504/IJHPCN.2019.102131

International Journal of High Performance Computing and Networking, 2019 Vol.14 No.3, pp.315 - 324

Received: 06 May 2017
Accepted: 02 Dec 2017
Published online: 09 Sep 2019 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article