Neural Graphs: an Effective Solution for the Resource Allocation in NFV Sites interconnected by Elastic Optical Networks

The paper proposes and investigates neural graph-based solution for the prediction of the processing capacities needed by the Virtual Network Function Instances in a Network Function Virtualization environment. The proposed solution is centralized and performed by the Orchestrator which: i) acquires the processing capacity values measured by the Virtual Network Function Manager; ii) builds neural graphs each one relative to a measure period and where each node of a graph represents a VNFI and it is labelled with the measured processing capacity of that VNFI; iii) evaluates the prediction of the processing capacities needed by each VNFI node by means of convolution operations that allow for a capture of spatial and temporal correlations of the processing capacities required by the VNFIs. Instead of applying regular convolutional and recurrent units, we formulate the problem on graphs and build the model with complete convolutional structures, which enable much faster training speed with fewer parameters.