I build high-performance storage and memory systems. I obtained my Ph.D. degree in 2023 from the Computer Science department of Rutgers University under the advisory of Prof. Sudarsun Kannan. Following the completion of my Ph.D., where my thesis work focused on building high-performance storage systems, I am excited to join École Polytechnique Fédérale de Lausanne (EPFL) as a postdoc researcher in RS3Lab in early 2024. My research interests will continue to revolve around building high-performance storage systems, with a particular emphasis on the intersection with memory systems, security, programming languages, and large language models (LLM).
News
30th Apr. 2024 Our paper Monarch: A Fuzzing Framework for Distributed File Systems is accepted in ATC'24
12th Mar. 2024 Our US Patent Systems and methods for heterogeneous storage systems is granted as US 11,928,336 B2
11th Dec. 2023 I will be joining EPFL as a postdoc fellow in early 2024.
09th Dec. 2023 Our paper OmniCache: Collaborative Caching for Near-storage Accelerators is accepted in FAST'24
06th Dec. 2023 Our paper LLM as OS, Agents as Apps: Envisioning AIOS, Agents and the AIOS-Agent Ecosystem is available on arXiv
28th Nov. 2023 I successfully defended my Ph.D. thesis with the title "Toward Scalable and High-Performance I/O with Cross-layered Storage Design"!
27th Feb. 2023 Our paper FusionFS: Fusing I/O Operations in Firmware File Systems is accepted in NVMW'23
2nd Feb. 2022 I will be serving as the web chair in HotStorage'22
10th Dec. 2021 Our paper FusionFS: Fusing I/O Operations in Firmware File Systems is accepted in FAST'22
10th Feb. 2021 Our paper CrossFS: A Cross-layered Direct-Access File System is selected as Memorable Paper Award Finalist in NVMW'21
29th Jan. 2021 Our paper CrossFS: A Cross-layered Direct-Access File System is accepted in NVMW'21
18th Nov. 2020 Our paper KLOCs: Kernel-Level Object Contexts for Heterogeneous Memory System is accepted in ASPLOS'21
15th Aug. 2020 Our paper CrossFS: A Cross-layered Direct-Access File System is accepted in OSDI'20
28th Apr. 2020 Our paper CompoundFS: Compounding I/O Operations in Firmware File Systems is accepted in HotStorage'20
2nd Sep. 2019 My research abstract and poster ParaFS: Supporting Concurrency with Direct Access File Systems got accepted in Student Research Competition in SOSP'19
1st Sep. 2019 Student Travel Grant for SOSP'19, Thanks ACM SIGOPS!
27th Aug. 2019 Successfully passed my Ph.D. Qualifying Exam with the topic of "Providing Concurrency in Direct Access File Systems", thanks for all the advices from exam committee!
A Cross-layered Direct Access File Systems
We have entered the micro-second era of storage devices with ultra-fast SSDs and byte-addressable NVMs where applications like databases and key-value stores frequently access storage devices with randoms and small updates across tens and even hundreds of cores. Such hardware and software trends motivates us to rethink the structure of traditional file systems and I/O subsystem in kernel (VFS, file system, page cache, BIO, Device driver) where OS trapping takes few micro-seconds. Inode-centric lock in state-of-the art file systems like ext4 lock the entire inode even if a process only modify parts of a file which impedes concurrency and scalability.
To reduce software overhead in I/O subsystem and provide near hardware latency as storage are becoming faster, we leverage the compute power in both host and storage devices.
To improve scalability on shared file access among threads/processes, we exploited the idea of file descriptor based concurrency instead of inode-centric lock. File descriptor is a unique data structure for different threads/processes even when they are openning a shared file. Each file descriptor has its own I/O queue so that two threads writes two non-overlapping regions of the same file, no lock is required. Our file system checks and sort conflict updates if the regions are overlapping according to timestamp counter.
To utilize relatively weak CPU in storage device, we designed a device-level I/O scheduler framework where the storage device are responsible for multiplexing sorted file descriptor I/O queues with limited number of device CPU cores with different scheduling policies. (i.e. Round-Robin, Deadline, Urgent-Aware)
Monarch: A Fuzzing Framework for Distributed File Systems
Tao Lyu, Liyi Zhang, Zhiyao Feng, Yueyang Pan, Yujie Ren, Meng Xu, Mathias Payer, Sanidhya Kashyap
The 2024 USENIX Annual Technical Conference (ATC' 24)
[paper]
[slide]
[code]
LLM as OS, Agents as Apps: Envisioning AIOS, Agents and the AIOS-Agent Ecosystem
Yingqiang Ge, Yujie Ren, Wenyue Hua, Shuyuan Xu, Juntao Tan, Yongfeng Zhang
arXiv:2312.03815
[paper]
OmniCache: Collaborative Caching for Near-storage Accelerators
Jian Zhang, Yujie Ren, Marie Nguyen, Changwoo Min, Sudarsun Kannan
The 22th USENIX Conference on File and Storage Technologies (FAST' 24)
[paper]
[slide]
[code]
FusionFS: Fusing I/O Operations in Firmware File Systems
Jian Zhang*, Yujie Ren*, Sudarsun Kannan (* Co-first author)
The 20th USENIX Conference on File and Storage Technologies (FAST' 22)
[paper]
[slide]
[code]
KLOCs: Kernel-Level Object Contexts for Heterogeneous Memory System
Sudarsun Kannan, Yujie Ren, Abhishek Bhattacharjee
The 26th ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS' 21)
[paper]
[slide]
[code]
CrossFS: A Cross-layered Direct-Access File System
Yujie Ren, Changwoo Min, Sudarsun Kannan
The 14th USENIX Symposium on Operating Systems Design and Implementation (OSDI' 20)
[paper]
[slide]
[code]
CompoundFS: Compounding I/O Operations in Firmware File Systems
Yujie Ren, Jian Zhang, Sudarsun Kannan
The 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage' 20)
[paper]
[slide]
