MIT Department of Electrical Engineering & Computer Science

DATE: MONDAY, APRIL 10, 1995

TIME: TALK AT 1:00
Refreshments at 12:45

PLACE: NE43-518

Application-Controlled File Caching and Prefetching

Pei Cao
Princeton University

As disk performance continues to lag behind microprocessors and memory systems, the file system is increasingly becoming the bottleneck for many applications. In this talk, I show that with appropriate mechanisms and policies, application-controlled file caching and prefetching can significantly improve the file I/O performance of applications.

I propose two-level file cache management: the kernel allocates physical pages to individual applications, and each application is responsible for deciding how to use its physical pages. I focus on three issues: a global allocation policy that allows applications to control their own cache replacement and at the same time maintains fair allocation of cache blocks among processes, a low overhead mechanism to implement application/kernel interaction in two-level file cache management, and integrated algorithms for caching and prefetching.

Implementation demonstrates that good application-chosen replacement strategies, combined with our kernel allocation policy, can reduce the number of block I/Os by up to 80% and the elapsed time by up to 45% for some applications. In addition, prefetching integrated with cache replacement and disk scheduling can further reduce these applications' elapsed times by up to 25%.

HOST: Prof. Barbara Liskov