yFS: a journaling file system design for handling large data sets with reduced seeking

Abstract

In recent years, disk seek times have not improved commensurately with CPU performance, memory system performance, and disk transfer rates. Furthermore, many modern applications are making increasing use of large files. Traditional file system designs are limited in how they address these two trends. We present the design of a file system called yFS that consciously reduces disk seeking and handles large files efficiently. yFS does this by using extent-based allocations in conjunction with three different disk inode formats geared towards small, medium, and large files. Directory traversals are made efficient by using the B*-tree structure. yFS also uses lightweight asynchronous journaling to handle metadata changes. We have implemented yFS on FreeBSD and evaluated it using a variety of benchmarks. Our experimental evaluations show that yFS performs considerably better than the Fast File System (FFS) with Soft Updates on FreeBSD. The performance gains are in the range from 20% to 82%.