Suppressing Noisy Neighbours in 5G networks: An end-to-end NFV-based framework to detect and suppress noisy neighbours

Abstract

The 'noisy neighbour problem' refers to situations arising in network function virtualization where one or more virtualized units (such as virtual machines or Docker containers) experience a degradation in performance due to the fact that some of the resources needed are occupied by other units on the same node. This degradation in performance could be caused due to several reasons including inefficient scheduling procedures or a lack of compute, memory or network resources. Due to the multivariate nature of such situations, detecting them is non-trivial and requires different techniques like machine-learning. A common way to optimize such scenarios is by means of virtual machine (VM) or container migration. However, the resources required for migration are limited. Furthermore, the migration process is computationally expensive and comes with longer latency. This paper proposes an algorithm to suppress noisy neighbours using a combination of dynamic CPU pinning (or CPU affinity) based on host processor utilization and load balancing based on dynamic network slicing. An end-to-end framework proposed in this paper detects and suppresses noisy neighbours leading to improvement in the overall system efficiency.