Edge-native cable access network with Quic transport over UDP

This paper introduces a novel architectural model for cable access networks that leverages an edge-native approach to relocate QUIC protocol termination from centralized cores to the access edge. The proposed model addresses the performance limitations of traditional DOCSIS-based architectures, particularly under asymmetric traffic conditions and real-time service demands. To evaluate the model viability, a series of controlled experiments were conducted using the FD.io Vector Packet Processing (VPP) framework, the vppecho benchmarking tool, and the CSIT continuous performance testing infrastructure. Tests were executed on 3-node ICX-100GE testbed with Intel Ice Lake processors and 100GE Intel E810CQ interfaces, using a fixed network frame size of 1280 bytes and a single CPU core per instance. Four QUIC-over-UDP scenarios were analyzed: a baseline configuration (1 client, 1 session) achieving 882.8 Mbps; a 10-client, single-session case with 625.3 Mbps; a single-client, 10-session case with 586.5 Mbps; and a 10-client, 10-session configuration yielding 562 Mbps. The results show that the baseline configuration offers the highest throughput, while scaled session and client loads introduce measurable performance degradation due to increased overhead. This degradation is attributed to session tracking overhead, NUMA effects, and per-client context switching in a constrained single-threaded environment, highlighting the importance of scaling strategies. Its flexibility and software-defined nature make it applicable to evolving requirements in low-latency and scalable broadband services. These findings validate the efficiency and scalability of user-space QUIC transport at the network edge. This work lays the foundation for further research into edge-hosted QUIC processing in distributed DOCSIS environments. The created edge-native model demonstrates its suitability for cost-effective stepwise modernization of optical-coaxial access networks, offering improved responsiveness and reduced reliance on centralized infrastructure without requiring physical plant replacement.