CNS Core:Small: File System Offloading

Computers use filesystems that facilitate applications to store and access data in a storage device. Filesystems reside inside a trusted Operating System (OS) and enable consistent, secure, and durable access to data. Applications run inside untrusted user space and must cross the trusted OS boundary for accessing storage hardware using filesystems. In this era of ultra-fast storage technologies such as flash and nonvolatile memory, crossing the OS boundary for input/output (I/O) access incurs high software overheads and impacts I/O and application performance. Therefore, reducing software overheads is critical but must be done without compromising data consistency, security, and durability. This project investigates the redesign of a high-performance filesystem that allows applications to access the storage hardware directly bypassing the OS, while respecting crash-consistency, security, and durability. The project involves three research thrusts. The first thrust identifies critical filesystem components responsible for crash-consistency, security, and durability and offloads them into the storage hardware allowing applications to bypass the OS. The second thrust explores the use of accelerators for accelerating offloaded components for higher performance. The third thrust redesigns offloaded components to exploit hardware capabilities such as device capacitance that can reduce software overheads and improve crash-consistency. By redesigning storage software, this project has the potential to significantly improve I/O performance for a wide domain of applications including Internet of Things (IoT) as well as personal, datacenter, and mission-critical systems where running commodity OS is not feasible. Equally, the ideas and prototypes developed will educate undergraduate, graduate, and high-school students on the benefits of reducing software overheads, understanding the hardware capabilities when designing software, and hardware-software codesign. To evaluate the filesystem design, the project builds a software prototype and an accelerated emulator running a wide-range of personal computing and datacenter applications. The research publications, source code with documentation, and the instructions for reproducing published measurement of the system will be made public and available from the Rutgers University's (https://www.cs.rutgers.edu/~sk2113/fsoffload) website maintained for the duration of the project. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.