摘要
针对网格主机负载的特性,本文提出基于基因表达式编程和滑动窗口法构造网格主机负载时间序列模型的方法,扩充美国Dinda所开发的RPS的实验环境以支持基因表达式编程模型,并用该系统对Dinda和O’Halloran在1997年和1998年对39台DECAlphaDUX主机进行抽样所获得的主机负载信息进行实验。结果表明,用本文所提出的方法所构造的预测模型同自动回归模型(AR)相比,具有更高的预测精确度。
This paper proposes a grid host load time series model based on gene expression programming and slide windows, according to the statistical properties of grid host load. This paper constructs a gene-expression-programming-model- based prediction system using the RPS toolkit developed by Dinda, uses the RPS system to evaluate predictive models utilizing measurement traces, which I）inda and O＇ Halloran obtained by sampling the 39 DEC Alpha DUX host in 1997 and 1998. The results of the statistical analysis indicate that the gene expression programming model is more accurate than the AR model in predictive power.
出处
《计算机工程与科学》
CSCD
2008年第8期105-107,共3页
Computer Engineering & Science
