Terra incognita: On the practicality of user-space file systems

Abstract

To speed up development and increase reliability the Microkernel approach advocated moving many OS services to user space. At that time, the main disadvantage of microkernels turned out to be their poor performance. In the last two decades, however, CPU and RAM technologies have improved significantly and researchers demonstrated that by carefully designing and implementing a microkernel its overhead can be reduced significantly. Storage devices often remain a major bottleneck in systems due to their relatively slow speed. Thus, user-space I/O services, such as file systems and block layer, might see significantly lower relative overhead than any other OS services. In this paper we examine the reality of a partial return of the microkernel architecture-but for I/O subsystems only. We observed over 100 user-space file systems have been developed in recent years. However, performance analysis and careful design of user-space file systems were disproportionately overlooked by the storage community. Through extensive benchmarks we present Linux FUSE performance for several systems and 45 workloads. We establish that in many setups, FUSE already achieves acceptable performance but further research is needed for file systems to comfortably migrate to user space.