With the rise of single-digit microsecond latency storage devices, the software of the OS storage stack is becoming the primary bottleneck when accessing fast storage devices. We are designing a new framework, XRP (eXpress Resubmission Path), that uses allows storage applications to bypass the OS storage stack when submitting storage requests. XRP can lead to ~2X better latency and 2-3X better throughput for common operations, like index lookups and aggregations.
XRP: In-Kernel Storage Functions with eBPF, Yuhong Zhong, Haoyu Li, Yu Jian Wu, Ioannis Zarkadas, Jeffrey Tao, Evan Mesterhazy, Michael Makris, Junfeng Yang, Amy Tai, Ryan Stutsman, Asaf Cidon, Usenix OSDI 2022 (forthcoming)
BPF for Storage: An Exokernel-Inspired Approach, Yuhong Zhong*, Hongyi Wang*, Yu Jian Wu*, Asaf Cidon, Ryan Stutsman, Amy Tai, Junfeng Yang (* equal contribution), ACM HotOS 2021
Faculty / senior researchers: Asaf Cidon, Ryan Stutsman, Amy Tai, Junfeng Yang
PhD students: Yuhong Zhong, Haoyu Li, Ioannis Zarkadas, Ashfaqur Rahaman
MS students / undergrads: Tal Zussman, Evan Mesterhazy, Hongyi Wang, Michael Makris, Yu Jian Wu, Jeremy Carin
Fall 2021: class project in EECS 6897, distributed storage systems
Fall 2020: class project in EECS 6897, distributed storage systems