Flexible architecture for the future internet scalability of SDN control plane
Software-Defined Networking (SDN) separates the control plane from the data plane. The initial SDN approach involves a single centralized controller, which may not scale properly as a network grows in size. Distributed controllers have emerged to address the disadvantages of a single centralized con...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Estado: | Versión publicada |
| Fecha de publicación: | 2022 |
| País: | España |
| Institución: | CBUC, CESCA |
| Repositorio: | TDR. Tesis Doctorales en Red |
| OAI Identifier: | oai:www.tdx.cat:10803/673562 |
| Acceso en línea: | http://hdl.handle.net/10803/673562 https://dx.doi.org/10.5821/dissertation-2117-363398 |
| Access Level: | acceso abierto |
| Palabra clave: | Àrees temàtiques de la UPC::Informàtica 004 |
| Sumario: | Software-Defined Networking (SDN) separates the control plane from the data plane. The initial SDN approach involves a single centralized controller, which may not scale properly as a network grows in size. Distributed controllers have emerged to address the disadvantages of a single centralized controller. The control architecture needs to be distributed with traffic control between switches and controllers and among the controllers in order to allow SDNs for several thousand switches. One of the most significant research challenges for distributed controller architectures is to effectively manage controllers, which includes allocating enough controllers to appropriate network locations. To address these daunting issues, we make the following major contributions: This thesis expands the method of solving the Control Placement Problem (CPP) based on the K-means and K-center algorithms to include a Hierarchical Controller Placement Problem (HCPP), located at a high level of Super Controller (SC), a middle level of Master Controllers (MCs), and the lowest level of domain controllers (DCs). The optimization metric addresses latency between the controller and the switches assigned to it.. The proposed architecture and methodology are implemented using the topology of Western European NRENs from the Internet Topology Zoo. The entire network topology is divided into clusters, and the optimal number of controllers (DCs) and their placement are determined for each cluster. MC placement optimization determines the optimal number of MCs and their optimal placement. As a second contribution, an accumulated latency is defined to solve CPP, which takes into account both the latency between the controller and its associated switches and the latency between controllers. Under the constraint of latency, an optimization problem is formulated as per mixed-integer linear programming (MILP). The goal of the research is to reduce accumulated latency while also reducing the number of network controllers and optimizing their placement to achieve an optimal balance. The performance of the developed method is evaluated on Internet2 OS3E real network topology. To achieve the third objective, a metric was developed that includes reliability. The communication latency between controllers should also be considered because a low controller-switch delay does not always imply a short controller-controller delay for a particular controller placement. As the third contribution, we propose a novel metric for CPP to improve the reliability of controllers that takes into account both communication latency and communication reliability between switches and controllers, as well as between controllers. When a single link fails, reliability is taken into account. This aspect concluded by identifying the optimal controller placement to achieve low latencies in control plane traffic. The goal of this project is to reduce the average latency. As the fourth contribution, this study evaluates the Joint Latency and Reliability-aware Controller Placement (LRCP) optimization model. As the evaluation metric, control plane latency (CPL) is defined as the sum of the average switch-to-controller latency and average inter-controller latency. The latency of the control plane, utilizing the actual latencies of the real network topology, is calculated for every optimum placement in the network. In the case of a failure of the single link, the actual CPL for LRCP placements is calculated and evaluated to determine how good LRCP placements are. CPL metrics are used to compare latency and reliability metrics with other models. This study provides proof that the developed methodologies for large-scale networks are highly powerful in terms of searching for all feasible controller placements while assessing the outcomes. In addition, compared to previous work including latency among controllers and reliability for an event of single-link failure. |
|---|