Informed Prefetching and Caching (CMU-CS-97-204)

摘要: Disk arrays provide the raw storage throughput needed to balance rapidly increasing processor performance. Unfortunately, many important, I/O-intensive applications have serial I/O workloads that do not benefit from array parallelism. The performance of a single disk remains bottleneck on overall for these applications. In this dissertation, I present aggressive, proactive mechanisms tailor file-system resource management needs particular, will show how use applicationdisclosed access patterns (hints) expose and exploit parallelism, dynamically allocate file buffers among three competing demands: prefetching hinted blocks, caching blocks reuse, recently used data unhinted accesses. My approach estimates impact alternative buffer allocations application elapsed time applies run-time cost-benefit analysis where they greatest impact. implemented TIP, an informed manager, in Digital UNIX operating system measured its 175 MHz Alpha workstation equipped with up 10 disks running range Informed ten-disk reduces wall-clock computational physics, text search, scientific visualization, relational database queries, speech recognition, object linking by 10-84% average 63%. On disk, parallelism is unavailable avoiding accesses most beneficial, same 36% 13% compared alone. Moreover, applied multiprogrammed, workloads, TIP increases throughput.