CPU Resource Usage Analysis for Downlink PDCP Processing in CRAN

For the proper deployment of virtual Radio Access Network (RAN) functions in a cloud (i.e., server) environment, the resource requirements in terms of CPU, bandwidth and memory of these functions have to be evaluated. In this paper, we consider a distributed Cloud RAN (CRAN) architecture, where the PDCP protocol layer is executed at the so-called Central Unit (CU) (centralized cloud site), whereas the remaining RAN protocols (RLC, MAC and PHY) run in the Distributed Unit (DU), close to the antenna site due to their strict latency requirements. As the PDCP processing is quite resource consuming, we focus on the evaluation of its resource requirements. We assume that, in general, there are no midhaul and backhaul link capacity bottlenecks as well as no memory bottlenecks in the DU so that the most critical issue in CRAN is the PDCP processing resource requirement. Following this argument, we investigate the CPU resource consumption of the PDCP, considering its dependency on the packet length, the overall traffic load and the CPU clock rate. We perform measurements in a testbed using an open-source PDCP implementation (OAI from Eurecom) running on a standard Linux computer.