Towards Performance-Aware Management of P4-based Cloud Environments

The recent trend to deploy programmable packet processors in cloud environments enhances the packet processing capability without losing the flexibility to adapt the functions at runtime. However, managing network functions, in particular deciding where to instantiate a certain function, is a non-trivial task with many decisive factors. In this paper, we propose a mathematical model for optimizing the placement of network functions implemented in P4, considering the various types of devices with different properties in terms of processing delay and supported external functions and architectures. To model the processing delay, each network function is decomposed to a set of atomic constructs, whose latency has been properly measured. The numerical evaluation considering five types of network functions shows the effectiveness of the optimization model in selecting the number of devices to be used and in minimizing the overall packet forwarding delay and costs.