In this talk, we propose to satisfy these demands with a micro-file-system architecture (uFS). uFS consists of a set of modules that represent the fundamental subcomponents of a distributed file system and that are linked together via middleware that coordinates these modules to provide scalable, reliable file service to different applications. Like a micro-kernel, a micro-file-system seeks to separate mechanism from policy to simplify the design of the underlying system and to allow different policies to be instantiated to support different applications.

The uFS project is in its early stages at UT Austin. In this talk, we will outline some preliminary results that suggest that a uFS architecture will have three advantages over traditional file systems: more scalability, better reliability, and support for application-specific interfaces. Furthermore, we believe that a uFS represents a superior software engineering strategy compared to monolithic designs. We will also outline several of the key research questions that must be answered to make uFS a reality.

The uFS project is joint work with Lorenzo Alvisi and Harrick Vin.