Speaker: Tara M. Madhyastha, University of Illinois Urbana
Intelligent, Adaptive File System Policy SelectionDate: October 11, 1996
Abstract: Traditionally, maximizing input/output performance has required tailoring application input/output patterns to the idiosyncrasies of specific input/output systems. In the lecture, we show that one can achieve high application input/output performance via a low overhead input/output system that automatically recognizes file access patterns and adaptively modifies system policies to match application input/output needs. This approach reduces the application developer's input/output optimization effort by isolating input/output optimization decisions within a retargetable file system infrastructure.
To validate these claims, we have built a lightweight file system policy testbed that uses a trained learning mechanism to recognize access patterns. The file system then uses these access pattern classifications to select appropriate caching strategies, dynamically adapting file system policies to changing input/output demands throughout application execution. Our application domain includes low-level satellite data processing codes from the SeaWiFS and Pathfinder projects. Our experimental data show dramatic speedups on both benchmarks and these input/output intensive scientific applications.