MACHINE LEARNING METHOD FOR ADAPTIVE VIRTUAL NETWORK FUNCTIONS PLACEMENT AND READJUSTMENT

摘要

Virtual Network Functions (VNFs) are placed in a substrate network and the placement is readjusted based on dynamic resource availability and dynamic resource utilization in the substrate network. A predetermined number of servers is selected sequentially as cluster-heads based on a set of metrics which measure the efficiency of the servers in different aspects. The servers are partitioned into the predetermined number of disjoint clusters with different efficiency aspects. Each cluster includes one of the cluster-heads which performs the placement and readjustment of the VNFs for the cluster. An incoming VNF is placed at a given server in a given cluster by the cluster-head of the given cluster, which optimizes an objective function subject to a set of constraints. The objective function is optimized with respect to a subset of the metrics which excludes one or more metrics in which the given cluster is efficient.