Stop All File Systems Research

Abstract

We advocate that the OS community should stop all file systems research, because file systems are fundamentally the wrong place in the storage hierarchy to innovate. Three factors combine to limit the impact of file systems research: the constraints of legacy, the diversity of implementations, and a lack of information about storage-system internals. We instead suggest that researchers focus their innovations on storage systems, namely within disk or RAID subsystems. To enable a full range of functionality within the storage system, we posit that the disk system must be "semantically smart"; that is, it must understand the on-disk structures and recognize the on-line operations of the file system above. We discuss the concepts underlying semantically-smart disk systems, present a taxonomy of the different axes of semantic knowledge, and discuss the important remaining research challenges. 1 Background