Reproduce “OpenRDMA: A GPU-Native Multi-Path Reliable RDMA Transport Architecture” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Ziheng Ji, Chunyu Liu, Weihao Peng, Mingheng Mi, Pu Wang, Hai Zhang

OpenRDMA lets the GPU submit an 8-byte token for the NIC to reconstruct a preconfigured RDMA operation, while the remote NIC publishes a same-QP flag for GPU polling. In RTL/network co-simulation, packet spraying achieved 323.5 Gb/s payload goodput for one QP and 16-KiB messages, versus 92.2 Gb/s with fixed paths. Fixed mapping won at 4 KiB; results do not establish deployed-system performance.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.