高性能存储系统的读请求的建模和预测

为了改善高性能计算系统的I／0性能，本文设计和实现了一个读请求的空间和时间特性的综合预测模型-马尔可夫空间模型与自回归集成的移动平均时间模型，该模型在应用程序执行过程中被自动识别；每个新的请求到达时模型参数被实时递归地估计，并根据应用程序本身和运行环境自适应改变。经过实验检验我们的建模和预测比较有效，测试例子表明时间模型的预测值与观测值的相对误差的平均值为0．16％，并且预测的稳定性较好；空间模型的预测准确性也较好。

Automatic Modeling ,Forcasting of Read Request about High Performance Storage System

In order to improve I/O performance of high performance computing systems,this paper designs and implements an integration modeler to model and predict for read request interarrival times and spatial characteristic:models are automatically identified during application execution,model parameters are recursively estimated in real-time for every new I/O arrival,adapting to changes that are intrinsic to the running application,the implementation shows that our modeler has good veracity and validity,we compare and calculate 2 examples and know the average relative forecast error of interarrival times modeler is 0.16%,forecast is steady;and Markov model spatial predictions also has good veracity.