Optimisation of the Resilience of a Wireless Mesh Network for Disaster Scenarios through Network Functions Virtualisation with distributed Orchestration

Abstract

Wireless multi-radio outdoor-routers are building a multi-channel wireless mesh network (WMN), used as an emergency communication infrastructure for rescue-helpers and victims when a disaster destroys the existing communication infrastructure. The architecture for the WMN is based on a clustering concept, which defines a different wireless channel for each cluster. This provides a high throughput WMN with minimal interference for IP-based communication. The service provisioning within the network is achieved by integrating Network Functions Virtualisation (NFV) into the network. For this aspect, the hardware resources of the routers are utilised to establish the required NFV infrastructure (NFVI), which is used for providing services and network functionalities by deploying the corresponding virtual network functions (VNFs), such as a Multimedia over IP Call Server, a File Sharing Platform, Webservers, and corresponding servers for DNS (Domain Name System) and DHCP (Dynamic Host Configuration Protocol). Because it cannot be expected that the network has a connection to an external network, such as the Internet, it must work as a stand-alone and autonomous network and, therefore, requires the NFV management and orchestration (NFV-MANO) to operate within the network. The major issue of the WMN for disaster scenarios consists of dynamic changes influencing the availability and usability of the network and its services, namely node failures, node gains, occurring interferences and the movement of network end-users, which can occur at any given moment, including at the establishment of the network by the rescue-helpers deploying the hardware of the network. This thesis aims to eliminate the impact of a single point of failure resulting from a logically centralised NFV orchestrator (NFVO), the main component of a standardised NFV-MANO. By distributing the NFVO's functionality among the network nodes, the availability of the orchestration in the network shall be guaranteed. The primary task of the resulting distributed orchestration consists of optimising the network's resilience by ensuring the availability and usability of the network and its services in the face of the dynamic changes affecting the network infrastructure. This thesis proposes a distributed orchestration, a resilience management the orchestration is utilising, and a resilient placement concept for the network function required in the network. The distributed orchestration uses the clustered architecture of a multi-radio WMN for disaster scenarios. Each cluster in the network is assigned an instance of the distributed orchestration, namely a Cluster-Orchestrator, responsible for managing and orchestrating the resources in its assigned cluster. Through the communication between the existing Cluster-Orchestrators, a network-wide orchestration is achieved. The resilience management concept is based on monitoring the dependencies between the resources in the network, which supports the distributed orchestration in detecting, identi-fying, and remediating the impact of the dynamic changes within the configuration of the network. Furthermore, the resilient placement concept for the required network functions (i.e., Cluster-Orchestrators and VNFs, as well as their redundant instances) optimises their position within the network. The objective of the placement concept consists of minimising the probability of losing one or more network functions in case of a node failure. Simulations with a mathematical formulation of the optimisation show that the concept significantly reduces the probability of losing network functions due to different sizes of occurring network partitions compared to a pure performance- or random-based placement of the network functions. This contributes to the distributed orchestration in maintaining the network resilience due to a reduced workload required to recover the network after a node failure.