许多系统把数据访问请求当作是独立的事件。实际上,数据请求并非完全随机,而是由用户或程序的行为驱动的,不同的用户或程序存在不同的访问模式。LS(Last Successor)模型简单,但非常有效,然而它的预测结果严重依赖于用户或程序的访问顺...许多系统把数据访问请求当作是独立的事件。实际上,数据请求并非完全随机,而是由用户或程序的行为驱动的,不同的用户或程序存在不同的访问模式。LS(Last Successor)模型简单,但非常有效,然而它的预测结果严重依赖于用户或程序的访问顺序。提出了ULNS(User-based Last N Successors)文件预测模型,利用用户信息来提高预测精确度,并综合LS模型来改进算法的可适用度。实验结果表明,该预测模型具有较好的整体性能。展开更多
So far, file access prediction models is mainly based on either the file access frequency or the historical record of the latest access. In this paper, a new file access prediction model called frequency- and recency-...So far, file access prediction models is mainly based on either the file access frequency or the historical record of the latest access. In this paper, a new file access prediction model called frequency- and recency-based successor (FRS) is presented which combines the advantages of the file frequency with the historical record. FRS model has the capability of rapid response to workload changes and can predict future events with greater accuracy than most of other prediction models. To evaluate the performance of FRS mode, the Linux kernel is modified to predict and prefetch upcoming accesses. The experiment shows that FRS can accurately predict approximately 80% of all file access events, while maintaining an immediate successor queue (ISQ) per-file which only requires regular dynamic updates.展开更多
基金国家自然科学基金( the National Natural Science Foundation of China under Grant No.90412017)
文摘许多系统把数据访问请求当作是独立的事件。实际上,数据请求并非完全随机,而是由用户或程序的行为驱动的,不同的用户或程序存在不同的访问模式。LS(Last Successor)模型简单,但非常有效,然而它的预测结果严重依赖于用户或程序的访问顺序。提出了ULNS(User-based Last N Successors)文件预测模型,利用用户信息来提高预测精确度,并综合LS模型来改进算法的可适用度。实验结果表明,该预测模型具有较好的整体性能。
基金Supported by Key Technology R&D Project Foundation of Sichuan Province (No.02GG006-018)
文摘So far, file access prediction models is mainly based on either the file access frequency or the historical record of the latest access. In this paper, a new file access prediction model called frequency- and recency-based successor (FRS) is presented which combines the advantages of the file frequency with the historical record. FRS model has the capability of rapid response to workload changes and can predict future events with greater accuracy than most of other prediction models. To evaluate the performance of FRS mode, the Linux kernel is modified to predict and prefetch upcoming accesses. The experiment shows that FRS can accurately predict approximately 80% of all file access events, while maintaining an immediate successor queue (ISQ) per-file which only requires regular dynamic updates.
