µDCN:A High-Performance, ML-Orchestrated Data-CentricNetworking Architecture for Edge Environments

Abstract Edge computing environments require networkingarchitectures that can support heterogeneous devices, dynamicworkloads, and stringent latency constraints. Named Data Net-working (NDN) provides content-centric communication, in-network caching, and data-centric security, but existing imple-mentations often face a trade-off between performance, deploy-ability, and adaptability. This paper presents µDCN, a cross-layer NDN architecture for edge environments that integrates aneBPF/XDP fast path, a Rust-based QUIC transport layer, and aTensorFlow Lite control plane for ML-driven MTU adaptation.The proposed design exposes fine-grained telemetry from thepacket-processing path and uses it to dynamically tune packetgranularity under changing network conditions. A prototype isimplemented and evaluated on a heterogeneous edge testbedusing realistic traffic profiles and controlled link impairments.Results show that µDCN improves throughput and latency overrepresentative NDN implementations while reducing CPU andinference-engine resource consumption. The paper also analyzesdeployment constraints, security implications, and current limi-tations of the architecture.