Skip to content Skip to navigation

Acomplishment Based Renewal: FTFS: A Read/Write Optimized Fractal Tree File System

Thursday, September 12, 2019

Congratulations to Prof. Martin Farach-Colton for an Accomplishment-based Renewal of his NSF award "FTFS: A Read/Write Optimized Fractal Tree File System".

The research goal of this project is to demonstrate that it is possible to build a single, general-purpose file system such that any workload performs well on any storage device. Currently, a wide variety of end-users do not realize the full performance potential of their storage hardware, unless they are able to select a file system that is well-matched to each workload. The choice of file system can alter application performance by at least a factor of two, if not an order of magnitude, and there is no one file system that performs best on a range of workloads. If this work is successful, users will get optimal performance from any storage device on any workload.

Our work to date has made a general-purpose file system for hard disk drives (HDDs), called BetrFS, that roughly matches the performance of any other Linux file system for any workload, and significantly exceeds the performance of all other file systems in many cases. We believe that these results can be extended to a wider range of devices, both faster devices (SSDs and NVMe) and slower, complex devices (SMR and IMR), but will require additional data structure and systems work to get the most out of the devices. By modeling the principal performance characteristics, and exposing these parameters as tunable “knobs” at runtime, the file system can seamlessly adapt to different physical devices, even those not yet available. Our preliminary results also show that a file system based on write-optimized dictionaries, like BetrFS, can create fast file and directory clones that are highly time- and space-efficient. In order to unlock future research opportunities, we also propose significant work toward technology transfer. This includes both bringing BetrFS closer to deployability in production environments and making it easier to explore new directions in future research or by other research groups.

 

Referenced Faculty: