Evaluating the Impact of Network Function Virtualization (NFV) on End-to-End Latency

Abstract

Network Function Virtualization (NFV) has become an important facilitator of scalable, expendable, cost-efficient network designs through the decoupling of network functions with proprietary hardware and their implementation as software instances on virtualized platforms. Regardless of these benefits, there are concerns about the effect of NFV on the end-to-end (E2E) latency, especially when it comes to the latency-sensitive applications, including real-time communications, industrial automation, and edge computing. In this paper, the effect of NFV on E2E latency has been assessed by examining the underlying latency elements that have been presented by virtualization and service chaining. An exhaustive analysis of processing, queuing, transmission and delay caused by virtualization is also introduced to determine important latency bottlenecks. To create a realistic traffic and workload environment, an experimental NFV testbed is developed based on a virtualized cluster and NFV management and orchestration system. Several service chain implementations and traffic profiles are tested to quantify the variation of latency at varying system loads, and deployment plans. The findings prove that although NFV may introduce extra processing and context-switching overhead over traditional hardware-based implementations, smart service placement, service chaining optimization, and effective orchestration can go a long way to reduce latency-penalties. The analysis also shows trade-offs among flexibility and performance with the focus on the significance of lightweight virtualization and resource-aware orchestration. In general, this paper has offered useful information on the latency behavior of NFV-enabled networks and has also given recommendations on how to design low-latency virtualized network services that can be used in the next-generation communication systems.