面向人工智能场景的数据处理方法、装置及存储介质
By optimizing data transmission between the GPU server and the PFS system using RDMA and RoCEv2 protocols in a private cloud intelligent computing scenario, the high CPU consumption and latency issues caused by the traditional Ethernet protocol are resolved, achieving efficient and low-latency data processing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING BAIDU NETCOM SCI & TECH CO LTD
- Filing Date
- 2026-02-27
- Publication Date
- 2026-07-17
AI Technical Summary
In private cloud computing scenarios, data transmission between GPU servers and parallel file storage systems relies on traditional Ethernet protocols, resulting in high CPU computing resource consumption and large data transmission latency, which becomes a bottleneck for system performance.
By employing the Remote Direct Memory Access (RDMA) protocol combined with the RoCEv2 protocol, the data transmission path is optimized through a virtual switch and intelligent computing gateway. This bypasses the operating system kernel protocol stack and enables direct data transmission between the GPU server and the PFS system memory. RDMA packets are processed first, and address mapping and encapsulation are performed.
Significantly improves data processing efficiency, reduces CPU utilization, lowers data transmission latency, and enhances GPU computing task efficiency and overall performance.
Smart Images
Figure CN121750404B_ABST