AbstractThe escalation of communication demands and the emergence of new telecommunication concepts such as 5G cellular system and smart cities requires the consolidation of a flexible and manageable backbone network. These requirements motivated the researcher to come up with a new placement algorithm for the Controller of Virtual Network (COVN). This is because SDN and network virtualisation techniques (NFV and NV), are integrated to produce multiple virtual networks running on a single SD-WAN infrastructure, which serves the new backbone.
One of the significant challenges of SD-WAN is determining the number and the locations of its controllers to optimise the network latency and reliability. This problem is fairly investigated and solved by several controller placement algorithms where the focus is only on physical controllers.
The advent of the sliced SD-WAN produces a new challenge, which necessitates the SDWAN controllers (physical controller/hosted server) to run multiple instances of controllers (virtual controllers). Every virtual network is managed by its virtual controllers. This calls for an algorithm to determine the number and the positions of physical and virtual controllers of the multiple virtual SD-WANs. According to the literature review and to the best of the author knowledge, this problem is neither examined nor yet solved. To address this issue, the researcher designed a novel COVN placement algorithm to compute the controller placement of the physical controllers, then calculate the controller placement of every virtual SD-WAN independently, taking into consideration the controller placement of other virtual SD-WANs.
COVN placement does not partition the SD-WAN when placing the physical controllers, unlike all previous placement algorithms. Instead, it identifies the nodes of the optimal reliability and latency to all switches of the network. Then, it partitions every VN separately to create its independent controller placement. COVN placement optimises the reliability and the latency according to the desired weights. It also maintains the load balancing and the optimal resources utilisation. Moreover, it supports the recovering of the controller failure.
This novel algorithm is intensively evaluated using the produced COVN simulator and the developed Mininet emulator. The results indicate that COVN placement achieves the required optimisations mentioned above. Also, the implementations disclose that COVN placement can compute the controller placement for a large network ( 754 switches) in very small computation time (49.53 s). In addition, COVN placement is compared to POCO algorithm. The outcome reveals that COVN placement provides better reliability in about 30.76% and a bit higher latency in about 1.38%. Further, it surpasses POCO by constructing the balanced clusters according to the switch loads and offering the more efficient placement to recover controller-failure.
|Date of Award||Oct 2018|
|Supervisor||Ali Al-Sherbaz (Supervisor), James Xue (Supervisor) & Scott Turner (Supervisor)|