作者位次 #6/15OSDI 2026Oral· Operating Systems & Distributed Systems
UCCL-Tran: An Extensible Software Transport Layer for GPU Networking
Yang Zhou · Zhongjie Chen · Ziming Mao · ChonLam Lao · Shuo Yang · Pravein Govindan Kannan · Xizhi Zhang · Jiaqi Gao · Yilong Zhao · Yongji Wu · Kaichao You · Fengyuan Ren · Zhiying Xu · Costin Raiciu · Ion Stoica
摘要
Fast-evolving machine learning (ML) workloads have increasing requirements for networking. However, host network transport on RDMA NICs is hard to evolve, causing problems for ML workloads. For example, single-path RDMA traffic is prone to flow collisions that severely degrade collective communic...