读请求的空间与时间特征建模

Automatic temporal and spatial modeling of read request

为了改善并行应用程序和并行文件系统的I/O性能,对读请求的空间与时间特征进行建模。使用自相关函数分析和Haar小波变换来自动识别自回归、集成的移动平均的模型结构,通过该模型来预测读请求的时间特征;使用马尔可夫模型对读请求的空间特征进行建模、预测。该模型可以将自回归、集成的移动平均时间预测模型与马尔可夫空间预测模型结合,并自适应地预测什么时间、取哪些数据块、取多少数据块。

In order to improve I/O performance of parallel applications and parallel file systems, an automatic modeler was presented to model and predict for read request temporal and spatial series. It used aatocorrelation function and Haar wavelet transform technology to automatically identify and build Autoregressive Integrated Moving Average （ARIMA） models of interarrival times. It used Markov modeler to model and forecast spatial series. Our prefetching method can combine ARIMA time series predictions and Markov model spatial predictions to adaptively determine when, what, and how many data blocks to be prefetched.