The appearance of new network services and the ever-increasing network traffic and number of connected devices will push the evolution of current communication networks towards the Future Internet.
In the area of optical networks, wavelength routed optical networks (WRONs) are evolving to elastic optical networks (EONs) in which, thanks to the use of OFDM or Nyquist WDM, it is possible to create super-channels with custom-size bandwidth. The basic element in these networks is the lightpath, i.e., all-optical circuits between two network nodes. The establishment of lightpaths requires the selection of the route that they will follow and the portion of the spectrum to be used in order to carry the requested trac from the source to the destination node. That problem is known as the Routing and Spectrum Assignment (RSA) problem, and new algorithms must be proposed to address this design problem.
Some early studies on elastic optical networks studied gridless scenarios, in which a slice of spectrum of variable size is assigned to a request. However, the most common approach to the spectrum allocation is to divide the spectrum into slots of fixed width and allocate multiple, consecutive spectrum slots to each lightpath, depending on the requested bandwidth. Moreover, EONs also allow the proposal of more flexible routing and spectrum assignment techniques, like the split-spectrum approach in which the request is divided into multiple "sub-lightpaths".
In this thesis, four RSA algorithms are proposed combining two different levels of flexibility with the well-known kshortest path and first fit heuristics. After comparing the performance of those methods, a novel spectrum assignment technique, Best Gap, is proposed to overcome the inefficiencies emerged when combining the first fit heuristic with highly flexible networks. A simulation study is presented to demonstrate that, thanks to the use of Best Gap, EONs can exploit the network flexibility and reduce the blocking rate.
On the other hand, operators must face profound architectural changes to increase the adaptability and flexibility of networks and ease their management. Thanks to the use of network function virtualisation (NFV), the necessary network functions that must be applied to o er a service can be deployed as virtual appliances hosted by commodity servers, which can be located in data centres, network nodes or even end-user premises. The appearance of new computation and networking paradigms, like multi-access edge computing (MEC), may facilitate the adaptation of communication networks to the new demands. Furthermore, the use of MEC technology will enable the possibility of installing those virtual network functions (VNFs) not only at Data Centres (DCs) and Central Oces (COs), traditional hosts of VFNs, but also at the edge nodes of the network. Since data processing is performed closer to the enduser, the latency associated to each service connection request can be reduced. MEC nodes will be usually connected between them and with the DCs and COs by optical networks.
In such a scenario, deploying a network service requires completing two phases: the VNF-placement, i.e., deciding the number and location of VNFs, and the VNF-chaining, i.e., connecting the VNFs that the trac associated to a service must transverse in order to establish the connection. In the chaining process, not only the existence of VNFs with available processing capacity, but the availability of network resources must be taken into account to avoid the rejection of the connection request. Taking into consideration that the backhaul of this scenario will be usually based on WRONs or EONs, it is necessary to design the virtual topology (i.e., the set of lightpaths established in the networks) in order to transport the traffic from one node to another. The process of designing the virtual topology includes deciding the number of connections or lightpaths, allocating them a route and spectral resources, and finally grooming the traffic into the created lightpaths.
Lastly, a failure in the equipment of a node in an NFV environment can cause the disruption of the SCs traversing the node. This can cause the loss of huge amounts of data and affect thousands of end-users. In consequence, it is key to provide the network with fault management techniques able to guarantee the resilience of the established connections when a node fails.
For the mentioned reasons, it is necessary to design orchestration algorithms which solve the VNF-placement, chaining and network resource allocation problems in 5G networks with optical backhaul. Moreover, some versions of those algorithms must also implements protection techniques to guarantee the resilience system in case of failure.
This thesis makes contribution in that line. Firstly, a genetic algorithm is proposed to solve the VNF-placement and VNF-chaining problems in a 5G network with optical backhaul based on star topology: GASM (genetic algorithm for effective service mapping). Then, we propose a modification of that algorithm in order to be applied to dynamic scenarios in which the reconfiguration of the planning is allowed. Furthermore, we enhanced the modified algorithm to include a learning step, with the objective of improving the performance of the algorithm.
In this thesis, we also propose an algorithm to solve not only the VNF-placement and VNF-chaining problems but also the design of the virtual topology, considering that a WRON is deployed as the backhaul network connecting MEC nodes and CO. Moreover, a version including individual VNF protection against node failure has been also proposed and the effect of using shared/dedicated and end-to-end SC/individual VNF protection schemes are also analysed.
Finally, a new algorithm that solves the VNF-placement and chaining problems and the virtual topology design implementing a new chaining technique is also proposed. Its corresponding versions implementing individual VNF protection are also presented. Furthermore, since the method works with any type of WDM mesh topologies, a technoeconomic study is presented to compare the effect of using different network topologies in both the network performance and cost.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados