摘要：随着5G/6G通信、边缘计算与人工智能技术的深度融合，传统网络架构面临算力资源分布不均、服务响应延迟高、动态适应能力不足等挑战。提出三级分层在网计算体系架构，通过算力泛在化、服务动态化与决策智能化的深度协同，解决算网协同关键瓶颈。该架构在数据中心网络速率限制中显著提升处理性能与系统稳定性，为高效数据传输、灵活网络服务及大规模智能训练提供支撑，推动网络向高性能、智能化方向演进。

关键词：在网计算；架构；算网协同；智能网络技术

Abstract: With the deep integration of 5G/6G communication, edge computing, and artificial intelligence technologies, the traditional network architecture is faced with challenges such as uneven distribution of computing power resources, high service response latency, and insufficient dynamic adaptation capabilities. A three-level hierarchical in-network computing architecture is proposed. Through the deep collaboration of ubiquitous computing power, dynamic services, and intelligent decision-making, the key bottleneck of computing-network collaboration is addressed. This architecture significantly improves the processing performance and system stability in the rate limiting of data center networks, providing support for efficient data transmission, flexible network services, and large-scale intelligent training, and promoting the evolution of the network towards high performance and intelligence.

Keywords: in-network computing; architecture; computing-network collaboration; intelligent network technology