Search papers, labs, and topics across Lattice.
This paper introduces MRC, a new RDMA-based transport protocol that sprays traffic across multiple paths and actively load-balances to mitigate flow collisions in large-scale AI training clusters. The authors combine MRC with multi-plane Clos topologies and static source-routing using SRv6 to enhance network resilience and redundancy. Production deployments at OpenAI and Microsoft demonstrate that MRC enables AI training jobs to withstand numerous network failures that would previously have caused interruptions.
AI training jobs can now shrug off network failures that used to halt progress, thanks to a new resilient networking stack deployed at OpenAI and Microsoft.
Tail latency dominates the performance of synchronous pretraining jobs when running at very large scales. We describe a three-pronged approach: (1) a new RDMA-based transport protocol, MRC, sprays across many paths and actively load-balances between them, eliminating the issue of flow collisions (2) the use of multi-plane Clos topologies to get the benefits of high switch radix and redundancy, allowing training clusters well over 100K GPUs to be built as two-tier topologies while increasing physical redundancy, and (3) the use of static source-routing using SRv6 to allow MRC the freedom to bypass failures by itself. We describe our experiences running MRC and static SRv6 routing in production in OpenAI and Microsoft's largest training clusters, where it has been used to train the latest frontier models. We demonstrate how MRC allows AI training jobs to ride out many network failures that previously would have interrupted training.
OpenAI