论文部分内容阅读
预取是提高存储系统性能的主要手段之一.但现有存储系统的设备层并不知道任何I/O访问的语义信息,因而不能充分利用I/O访问的语义来预取下一时刻要访问的数据,只能利用较简单的方式如I/O访问的局部性、顺序访问和循环访问等特性来实现简单的预测.为此,本文根据存储系统的特点提出了实用且高效的基于连续度的聚类算法来发现密集读请求访问的区域,并采用ARMA时间序列模型来预测密集读请求可能访问的区域及访问时刻,为正确的预取提供了准确的信息.为提高预取的准确性,并采用了动态参数估计的策略.通过大量实验的结果验证了这两种算法的正确性和预测的准确性,能较大的提高存储系统的预取效率.
Prefetching is one of the main ways to improve the performance of a storage system, but the device layer of the existing storage system does not know the semantic information of any I / O access and thus can not fully utilize the semantic of I / O access to prefetch the next moment Therefore, this paper proposes a practical and efficient method based on the characteristics of storage system, based on the characteristics of storage system, such as continuous, continuous, Degree clustering algorithm to find the area where intensive read requests are accessed and ARMA time series model is used to predict the areas that may be accessed by intensive read requests and the access time to provide accurate information for correct prefetching.In order to improve the accuracy of prefetching And adopts the strategy of dynamic parameter estimation.The results of a large number of experiments verify the correctness of these two algorithms and the accuracy of prediction and can greatly improve the prefetching efficiency of the storage system.