Performance analysis of VNFs for sizing cloud-RAN infrastructures

We explore in this paper the design of a Cloud-RAN architecture. We study the problem of determining the required computing capacity for hosting software-based Base Band Units (BBUs) of various eNodeBs in a data center. A worst-case analysis in terms of execution time of BBU functions is then performed. We specifically investigate how to reduce their runtime and thus to increase the concentration level of these functions in the back-haul network. For this purpose, we explore how to execute on a multi-core platform encoding and decoding tasks, which are the most greedy in terms of processing time. By introducing various parallelism schemes, we quantify the gain in latency. It turns out that parallelism on a Code Block (CB) basis is the most efficient as soon as the computing capacity of the multi-core platform is above a certain threshold. The gain is then substantial and scales with the processing capacity, allowing higher concentration of eNodeBs in the BBU-pool.