VNF modeling towards the cloud-RAN implementation

We address in this paper the problem of modeling Virtual Network Functions (VNFs) in order to devise execution principles. We specifically represent a VNF as a chain of sub-functions or micro-services and investigate its processing on a multi-core platform. Instead of reserving dedicated resources to a VNF, we specifically examine the gain that can be obtained by resource pooling, namely a pool of cores which is shared by active VNFs. The driving use case of this study is the virtualization and the cloudification of Radio Access Network functions, in particular Base Band Unit (BBU) functions. We validate this approach by simulation and show that an adequate threading model can reduce the VNF runtime. Obtained results open the door towards the Cloud-RAN implementation since some BBU tasks might be hosted in geographically distant data centers.