Improving Cache Performance by Predicting I/O System Actions

该项目正在通过根据过去发生的访问来预测未来的访问来寻找提高I/O性能的方法。结果可以应用于从磁盘控制器缓存到文件系统再到万维网的所有过程。例如，已经证明对文件的访问不是随机的。有一些模式可以被检测和利用。通过利用这些模式，可以预取文件，这样当请求发生时，文件就会出现在缓存中。使用文件系统跟踪的初步结果表明，性能将显著提高。该项目将(1)开发一种在空间和时间方面都有效的跟踪I/O访问模式的方法，(2)在磁盘控制器、文件系统和万维网这三个级别上调查预测性缓存预取的使用，以及(3)在这些领域中的一个或多个领域实施预测性缓存。***

This project is finding ways to improve I/O performance by predicting future accesses based on those that have occurred in the past. The results can be applied all the way from the disk controller cache to the file system to the world wide web. It has been demonstrated that accesses to files, for example, are not random. There are patterns that can be detected and exploited. By exploiting these patterns, files can be prefetched so that the files will be present in the cache when the request occurs. Preliminary results using file system traces indicate that significant performance gains will occur. This project will (1) develop a method for tracking I/O access patterns that is efficient in terms of both space and time, (2) investigate the use of predictive cache prefetching at three levels: disk controller, file system and the world-wide web, and (3) implement a predictive cache in one or more of these areas. ***