Towards an eBPF+XDP Based Framework for Open, Programmable and Scalable NextG RANs

Starting with 5G, radio access networks (RANs) are moving towards an disaggregated architecture, with most of its functionality (except for the low-level PHY layer) implemented in software. While software affords the benefits of programmability and scale-out, it is also far slower than hardware. This is further compounded by the needs for more complex, dynamic and intelligent features in Next-Generation (NextG) RANs. In this work, we advocate an eBPF (extended Berkeley Packet Fil-ter)+XDP (eXpress Data Path) based framework for scaling and accelerating software packet processing in (0- RAN compliant) NextG RANs. Using 5G Central Unit User Plane (CU-UP) as a key case study, we present an initial design of our proposed framework, dubbed PRANAVAM, and it's the key components. We also discuss additional options for further improvements. Our preliminary evaluation results shows that PRANAVAM improves the 5G CU-UP throughput by 22-26%, compared with the existing (open-source) 5G RAN implementation.