Robust Virtual Network Function Provisioning Under Random Failures on Network Function Enabled Nodes

摘要

Network function virtualization enables on-demand network function (NF) deployment providing agile and dynamic network services. Early works on NF focused on its provisioning, design, and management with quality metrics - NF-service availability and reliability under system failure(s). To facilitate flexible NF service recovery and migration with high reliability against random NF-enabled node failures, with a known NF resource pool, we first introduce a new NF service evaluation metric to quantify the minimum reliability among all requested NFs for all end-to-end demands - a tight lower bound on individual NF's service reliability among all requests. We then study the robust virtual network function (VNF) provisioning problem where only a limited number of VNF instances may be instantiated while maximizing the proposed evaluation metric. We present exact solution approach which guarantees the minimum reliability of all NF service to be in the range [76%, 94%] even when physical nodes may fail with a very high (50%) probability.