CSE PhD Student Wenxue LI (Advised by Prof. Kai CHEN) Received ACM SIGCOMM'25 Best Student Paper Award Honorable Mention

Miss Wenxue LI (advised by Prof. Kai CHEN), a PhD student from Department of Computer Science and Engineering, together with other collaborators, has received the prestigious Best Student Paper Award (Honorable Mention) from ACM SIGCOMM 2025 for their work "Revisiting RDMA Reliability for Lossy Fabrics."

Towards a more scalable and easy-to-operate datacenter network

Miss Wenxue Li expressed her sincere appreciation to her advisor, Prof. Kai Chen, for his years of mentorship in both technical research and broader academic training, as well as to all collaborators who contributed to this work. She emphasized that one key advancement in datacenter networking lies in continuously making networks more scalable and easier to operate.

"In addition to the intrinsic scalability challenges, lossless RDMA introduces significant operational risks such as workload interference and cascading failures. This can be a nightmare for network operation. By enabling RDMA over lossy fabrics, we eliminate these issues at the root, and that is why we are pursuing this direction," she explained.

She also noted that this recognition from SIGCOMM will further inspire her to carry out more solid research in this direction.

About the award-winning paper

The awarded paper addresses the fundamental challenge of shifting RDMA from lossless to lossy fabrics.

RDMA was originally designed for lossless InfiniBand networks, where RDMA NICs (RNICs) adopted a Go-Back-N retransmission mechanism to simplify processing. RNICs for Ethernet inherited this mechanism and relied on flow control to emulate a lossless Ethernet fabric. However, lossless RDMA introduces significant limitations, hindering scalability and large-scale deployment. To overcome these challenges, both academia and industry have been actively exploring efficient RDMA communication over lossy fabrics, though existing solutions still suffer from performance inefficiencies.

This paper proposes a new transport architecture, DCP, built on a switch–RNIC co-design paradigm that rethinks RDMA reliability. DCP improves loss recovery efficiency, is naturally compatible with packet-level load balancing, and remains feasible for hardware offloading. At its core, DCP separates the control plane (CP) for header transfer from the data plane (DP) for payload. While lossless RDMA enforces loss-free delivery for both planes, DCP guarantees a reliable CP while allowing the DP to operate over lossy fabrics. DCP then leverages the lossless CP feature to enhance the RNICs' reliability, enabling compatibility with packet-level LB, precise retransmission, and minimal memory and processing overhead.

DCP workflow

About ACM SIGCOMM

ACM SIGCOMM is the flagship annual conference of the ACM Special Interest Group on Data Communication. ACM SIGCOMM 2025, the 39th edition of the conference series held in Coimbra, Portugal on September 8-11, received 463 submissions and accepted 75 papers with around 16.2% acceptance ratio.

Miss Wenxue LI (third from left), the first author of the award-winning paper, began her Ph.D. studies in 2021 in the CSE department supervised by Prof. Kai CHEN (third from right)

Best Student Paper Award (Honorable Mention) - SIGCOMM 2025

Congratulations again to Miss Wenxue LI, Prof. Kai CHEN, and other co-authors on this outstanding achievement.