KDN-Based Fault-Tolerant Scheduling for VNFs in Data Centers

Abstract

Network Functions Virtualization (NFV) aims to decouple network functionality from dedicated physical devices, thereby allowing Virtual Machines (VMs) to host services such as firewalls and load balancers. Hosting these services on VMs will provide new business opportunities for data centers. Currently, the available physical resources are limited, thus making efficient resource scheduling the most challenging issue towards the successful adoption of NFV. Therefore, optimizing resource allocation is crucial. This paper proposes an integrated, intelligent framework for VNF (Virtual Network Function) scheduling. In particular, the proposed solution is based on Knowledge Defined Networking (KDN) with Long Short-Term Memory (LSTM) to operate and optimize VNF scheduling. For this reason, our solution considers VNF affinity rules in a data center with LSTM- based multi-step ahead time forecast. Experimental analysis through benchmark data shows that the proposed fault-tolerant aware scheduling framework based on LSTM and affinity techniques achieves promising results.