Rethinking File Caching for Flexible Address Space

Structured files used with database management systems are at the forefront of sophisticated analytic tools for data mining. The use of a flexible file address space beneath these structured files can achieve high performance read and writes. However, because FlexFile allows in-place updates to arbitrary-sized data, a new caching system must be realized which leverages this uniquely shifting logical address space. In this paper, we present the cache structures block caching and replacement policy as well as the process to integrate the cache to the Linux kernel via a modularized driver. To validate our methodologies, we measure our cache against the Linux Page Cache. Testing results show that certain cases where utilizing a logical cache can achieve better results over the Linux Page Cache.