Abstract:
The architecture of integrating Software Defined Networking (SDN) with Network Function Virtualization (NFV) is an excellent architecture because the former virtu- alizes the control plane and the latter virtualizes the data plane. As Programming Protocol-independent Packet Processors (P4) becomes popular, the architecture inte- grating SDN with NFV may be changed from traditional switches to P4 switches. In this architecture which integrates the P4 switch and NFV (P4+NFV), network func- tions can be provided in both P4 switches (PNF) and NFV (VNF). Thus, to minimize the packet delay, an offloading problem between P4 switches and NFV in this P4+NFV should be addressed. This work addresses this offloading problem and figures out the prioritization mechanism between newly coming packets and packets which require VNF for minimizing packet delay.
We model and analyze P4+NFV architecture using an M/M/1 queuing model with non-preemptive priority. This work also investigates the effect of using multiple VNF servers, which have different computing capacities, in P4+NFV architecture. Also, we propose an optimization solution based on gradient descent to find the optimal offload- ing probability of going to VNF/ VNFs.
Results show that optimal offloading from the P4 switch to NFV can reduce the average packet delay from 13.40% to 40.73%, when packets requiring VNF are given higher priority than the newly coming packets. Another findings is that optimal offload- ing from P4 switch to multiple VNF servers can reduce the average packet delay from 4.76% to 40.02%. Also, a larger capacity owned by a VNF server generates a higher offloading probability because sending more packets to it becomes more beneficial.
