An effective data aggregation based adaptive long term CPU load predictions mechanism on computational grid

RIS ID

38226

Publication Details

Dong, F., Luo, J., Song, A., Cao, J. & Shen, J. (2012). An effective data aggregation based adaptive long term CPU load predictions mechanism on computational grid. Future Generation Computer Systems: the international journal of grid computing: theory, methods and applications, 28 (7), 1030-1044.

Abstract

With the development of Internet-based technologies and the rapid growth of scientific computing applications, Grid computing becomes more and more attractive. Generally, the execution time of a CPU-intensive task on a certain resource is tightly related to the CPU load on this resource. In order to estimate the task execution time more accurately to achieve an effective task scheduling, it is significant to make an effective long-term load prediction in dynamic Grid environments. Nevertheless, as the prediction errors will be gradually accumulated while the best values of prediction parameters may vary vigorously, the existing prediction algorithms usually fail to achieve good prediction accuracy in the long-term prediction. To address these problems, an effective Data Aggregation based Adaptive Long term resource load Point-Prediction mechanism (DA2LPPoint) is proposed in this paper, where a data aggregation concept is introduced herein to reduce the number of prediction step. Furthermore, an interval based prediction mechanism with probability distribution representation called DA2LPInterval is lately proposed to improve the adaptation of prediction results. The experimental results show that the DA2LPPoint algorithm can outperform previous prediction methods in regard to mean square error (MSE). In addition, the DA2LPInterval algorithm can attain lesser prediction error with stronger representation capability; therefore, it is able to provide much more useful information for task scheduling in Grid environments.

Please refer to publisher version or contact your library.

Share

COinS
 

Link to publisher version (DOI)

http://dx.doi.org/10.1016/j.future.2011.10.014