Abstract

File systems have been mostly benchmarked as per the application perspective. This approach hides all the underlying complexities of the system including the actual I/O being done with the secondary storage device like magnetic disk. The IO bound property of a file system is necessarily to be evaluated because the most dominant performance limiting factor of a file system is its I/O operation with secondary storage device. This IO bound property of file system dictates the quantity and frequency of IO that a file system does with secondary storage device. In this paper, we argue system perspective of file system benchmarks and develop a benchmark to evaluate some common disk file systems for IO bound property. The goal of this paper is to better understand the behavior of file systems and unveil the low level complexities faced by file systems.