Disk Cache Architecture

在过去的五年中，随着微处理器性能和内存容量的提高，高性能超级计算机和“杀手级微型计算机”之间的主要区别之一是I/O性能。本研究的主要目标是通过提高磁盘缓存性能来提高I/O子系统的性能。考虑的第一个任务是通过分析I/O性能来描述呈现给磁盘控制器的负载。跟踪将通过检测磁盘控制器的设备驱动程序来收集。跟踪条目将被打上时间戳，因此可以获得系统负载的动态视图。调查中的第二个任务是使用跟踪驱动的模拟来调查各种磁盘缓存架构的性能。这项任务的一个主要目标是找到利用磁盘访问延迟的位置依赖特性的方法。因此，该项目将侧重于使用利用磁盘状态知识的缓存策略。

With the increases in microprocessor performance and memory capacity that have occurred during the past five years, one of the major difference between high performance supercomputers and the "killer micros" is I/O performance. The primary objective of this investigation is to increase the performance of the I/O subsystem by increasing disk cache performance. The first task considered is to characterize the load presented to a disk controller by analyzing I/O performance. The traces will be collected by instrumenting the device driver for the disk controller. Trace entries will be timestamped so a dynamic view of the system load can be obtained. The second task in the investigation is to use trace-driven simulation to investigate the performance of various disk cache architectures. A major goal of this task is to find ways of exploiting the position dependent nature of disk access latency. Thus, the project will focus on using cache policies that utilize knowledge of the disk's state.