SDR-RDMA: Software-Defined Reliability Architecture for Planetary Scale RDMA Communication
RDMA is vital for efficient distributed training across datacenters, but millisecond-scale latencies complicate the design of its reliability layer. We show that depending on long-haul link characteristics, such as drop rate, distance and bandwidth, the widely used Selective Repeat algorithm can be inefficient, warranting alternatives like Erasure Coding. To enable such alternatives on existing hardware, we propose SDR-RDMA, a software-defined reliability stack for RDMA. Its core is a lightweight SDR SDK that extends standard point-to-point RDMA semantics - fundamental to AI networking stacks - with a receive buffer bitmap. SDR bitmap enables partial message completion to let applications implement custom reliability schemes tailored to specific deployments, while preserving zero-copy RDMA benefits. By offloading the SDR backend to NVIDIA's Data Path Accelerator (DPA), we achieve line-rate performance, enabling efficient inter-datacenter communication and advancing reliability innovation for inter-datacenter training.
Authors
- Marcin Chrapek (ORCID: https://orcid.org/0009-0009-7654-6038)
- Sreevatsa Anantharamu (ORCID: https://orcid.org/0009-0006-0414-2200)
- Zhuolong Yu (ORCID: https://orcid.org/0000-0002-8846-5229)
- Tiancheng Chen (ORCID: https://orcid.org/0009-0002-8071-2552)
- Peter-Jan Gootzen (ORCID: https://orcid.org/0009-0008-8809-254X)
- Salvatore Di Girolamo (ORCID: https://orcid.org/0000-0003-2197-8860)
- Torsten Hoefler (ORCID: https://orcid.org/0000-0002-1333-9797)
- Mahmoud Elhaddad (ORCID: https://orcid.org/0009-0000-0335-7898)
- Abdul Kabbani (ORCID: https://orcid.org/0000-0001-7106-3824)
- Gil Bloch (ORCID: https://orcid.org/0009-0004-6224-9802)
- Jie Zhang (ORCID: https://orcid.org/0000-0002-5223-8446)
- Mikhail Khalilov (ORCID: https://orcid.org/0000-0003-0862-4662)
- Konstantin Taranov (ORCID: https://orcid.org/0000-0003-1356-5951)
- Scott Moe (ORCID: https://orcid.org/0009-0007-8500-5810)
- Joaquin José (ORCID: https://orcid.org/0000-0001-9549-7918)
- Rami Nudelman (ORCID: https://orcid.org/0009-0002-3601-2786)
- Siyuan Shen (ORCID: https://orcid.org/0009-0003-0061-5072)
- Kenji Nakano (ORCID: https://orcid.org/0009-0000-7701-8258)
- Nicola Mazzoletti (ORCID: https://orcid.org/0009-0000-6881-8261)
Institutions
- Microsoft (United States) (US)
- Nvidia (United Kingdom) (GB)
- Tamedia (Switzerland) (CH)
- ETH Zurich (CH)
- Microsoft (Finland) (FI)
- Wavin (Netherlands) (NL)
- Nvidia (United States) (US)
- CSCS - Swiss National Supercomputing Centre
- Swisscom (Switzerland) (CH)
Publication Details
- Published
- 2025-11-12
- DOI
- https://doi.org/10.1145/3712285.3759902
- Citations
- 1
- Primary Topic
- Cloud Computing and Resource Management
- Type
- article
- Field-Weighted Citation Impact
- 3.25