FullSight: a Deep Learning based Collaborated Failure Detection Framework of Service Function Chain

Abstract

Network Function Virtualization (NFV) is one of the most promising technologies which decouples Network Functions (NFs) from hardware resources to support more flexible network services and network resource allocation. However, these benefits increase the possibility of Service Function Chain (SFC) failures due to hardware failures, software defects and burst traffic, resulting in serious consequences. Unfortunately, existing failure detection methods have several issues, such as simplification of detection functionality, heavy overhead, and low accuracy. This paper introduces a framework FullSight, in which control plane and the programmable data plane can collaboratively detect failure and Deep Learning (DL) based algorithms are adopted for failure detection. FullSight can achieve an all-round perception of the state of the SFC, in which network information is acquired through the data plane, SFC components' message is obtained through the control plane. In addition, a failure detection model based on DL is established. Compared with the state-of-the-art methods, FullSight can support 8 kinds of the fine-grained failure detection. Our comprehensive evaluation of prototypes and simulations shows that FullSight can realize rapid and accurate detection and classification of diversified failures in SFCs. The bandwidth overhead reduces by 57% compared with the existing methods. Additionally, FullSight has a detection accuracy up to 93.5%.