OpenURMA: Open-Source Implementation of Huawei's Unified Bus Protocol
OpenURMA is the first clean-room open implementation of Huawei's Unified Bus (UB) protocol, a 2025 specification that redefines RDMA for datacenter networks. Traditional RDMA protocols like RoCE and InfiniBand suffer from per-connection state overhead, requiring hundreds of megabytes for 1024-application fanout and incurring four PCIe round trips per 64-byte operation, inflating latency. UB decouples per-application endpoint state from per-host transport state, making connection context additive, exposes ordering as opt-in, and enables remote memory access via native CPU load/store to an on-chip-bus controller. UB is currently implemented in Huawei's closed Ascend 950 silicon. OpenURMA implements UB's transport and transaction layers at three tiers, including synthesizable RTL on Alveo U. This open implementation aims to provide a royalty-free alternative for high-performance datacenter networking.
Key facts
- OpenURMA is the first clean-room open implementation of Huawei's Unified Bus (UB) protocol.
- UB is a public 2025 specification that changes the RDMA abstraction for datacenter networks.
- Traditional RDMA protocols like RoCE and InfiniBand hold per-connection state for every (application, remote-endpoint) pair.
- At 1024-application fanout, traditional RDMA requires hundreds of megabytes of state.
- Traditional RDMA pays a four-traversal PCIe round trip on a 64-byte operation, inflating latency an order of magnitude beyond the wire.
- UB decouples per-application endpoint state from per-host transport state so connection context grows additively.
- UB exposes ordering as opt-in and reaches remote memory through native CPU load/store to an on-chip-bus controller.
- UB ships in Huawei's closed Ascend 950 silicon.
- OpenURMA implements UB's transport and transaction layers at three tiers.
- OpenURMA includes synthesizable RTL on Alveo U.
Entities
Institutions
- Huawei
- OpenURMA