GSDM: Graph-Based Scaling Detection Model in Network Function Virtualization

Network function virtualization (NFV) is an emerging technology, which aims at replacing proprietary network function hardware devices with software- based network function (NF) applications. In NFV, it is critical to conduct dynamic and elastic resource allocation in accordance with varying workloads. State- of-the-art scaling detection algorithms are designed based on either traffic rate or runtime status. However, they cannot make precise and agile scaling decisions due to diversity of input traffic and noisy measurements. In this paper, we propose a novel graph-based scaling detection model (GSDM) using deep learning techniques. GSDM is a neural network model, which selects scaling action based on feature sequences of each NF and its adjacent NFs. Neural network provides a scalable way to incorporate both traffic rate and runtime status into control policy. Furthermore, adjacent NFs' feature information is also injected. As a result, GSDM empirically learns policies that achieves precise and agile scaling detections and improves system performance. Our implemented prototype demonstrates that GSDM achieves superior performance in precision and recall, and outperforms other schemes in a variety of network conditions.