算力智联网关键技术研究

摘要：智算业务的快速发展，计算资源需求急剧增长，对网络能力提出了更高标准。聚焦算力智联网，深入剖析数据入算、模型训练、模型推理三大典型智算业务场景需求，提出算力智联网关键技术要求。通过构建高效、灵活且可靠的智算基础设施，为智算业务提供差异化承载、灵活算网调度和无损传输等核心能力。此外，针对三大典型场景开展网络创新能力试点验证，显著提升了人工智能模型训练效率。

关键词：算力智联网；差异化承载；灵活算网调度；无损传输

Abstract: The rapid development of intelligent computing services has led to a sharp increase in computing resource demands, imposing higher standards on network capabilities. Focusing on the intelligent computing power network, the requirements of three typical intelligent computing business scenarios are analyzed: data ingestion, model training, and model inference. And the key technical requirements of the intelligent computing power network are put forward. By building an efficient, flexible, and reliable intelligent computing infrastructure, it provides core capabilities such as differentiated network bearer, flexible computing network scheduling, and lossless transmission for intelligent computing services. In addition, pilot verification work of network innovation capabilities has been carried out for the three typical scenarios, which has significantly improved the training efficiency of artificial intelligence models.

Keywords: intelligent computing power network, differentiated network bearing, flexible computing and network scheduling, lossless transmission