Contributions to routing scalability and qos assurance in cloud data transport networks based on the recursive internetwork architecture
- Autores: Sergio Leon Gaixas
- Directores de la Tesis: Jordi Perelló Muntan (dir. tes.), Davide Careglio (codir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2018
- Idioma: español
- Tribunal Calificador de la Tesis: Josep Lluís Marzo i Lázaro (presid.), Jordi Domingo i Pascual (secret.), David Larrabeiti López (voc.)
- Programa de doctorado: Programa de Doctorado en Arquitectura de Computadores por la Universidad Politécnica de Catalunya
- Materias:
- Enlaces
- Tesis en acceso abierto en: TDX
- Resumen
With an increasing number of devices and heterogeneous distributed applications, it is becoming evident that service delivered by the current Internet fall short to supply the actual Quality of Service (QoS) requirements of applications. In addition, the global scope of the IP layer causes large scalability problems on the network. Multiple solutions aim to overcome the limitations of the model (BGP, NAT, etc.), but all end being constrained by the same networking model that they try to improve, ending simply breaking and patching the stack itself of TCP/IP. In contrast, RINA proposes a new clean-slate Internet architecture based on a recursive networking stack with focus on inter process communication, where each layer, or DIF, performs the same set of tasks. DIFs are fully configurable by mean of programmable policies, and provide complete support for QoS services. RINA is capable to provide a standardized way to express the capabilities of each layer, the QoS Cubes. With those, RINA allows for applications and upper processes to express their requirements in terms of latency, losses, etc. The contributions in this thesis take profit from the recursive stack of RINA and the use of policies to propose and analyse old and new solutions which would not be compatible with the current TCP/IP Internet.
Improving the QoS services, this work takes profit from the information on requirements provided by the applications themselves to improve the assurance of QoS. With the use of ¿Q-based scheduling policies, improved QoS assurances are provided, aiming to provide “good enough” service for all flows in the network, resulting in a more appropriate sharing of resources. These policies have been tested in backbone-like networks, showing interesting improvements with respect to commonly used solutions like MPLS-based VPNs. In addition the provisioning of QoS services to end-users is also considered. In order to allow that, it is required to impose some limits on what end-users can send to the network, limiting the amount of priority traffic that potentially greedy users can send. In that regard, while enforcing strict rate-limits per QoS would be trivial in RINA, a new ¿Q-based rate-limiting policy that aims to limit the amount of priority traffic in a more user-friendly way is also explored.
In terms of scalability, this work also considers different measures to improve forwarding and routing within large-scale networks. As for the use of policies that could profit from specific network topologies, a new forwarding policy, that mix both topological rules and exceptions, is proposed. With this policy, forwarding table lookups in large tables are replaced with fast and simple forwarding rules based on the location of nodes and their neighbourhood. Given the common topologies used in large data centres, the proposed policy is found to be a perfect match for those scenarios. Test for different data centre topology showed clear improvements, requiring only a small fraction of all forwarding information despite the large size of such networks, depending that in the number of concurrent failures in the network rather than on the size of it. In addition, this work also considers the use of topological routing policies to populate exceptions upon failures. The use of topological routing solutions resulted in reduced complexity for computing paths and less routing messages. In addition to topological solutions, the use other routing solution, not well suited for the IP environment are also investigated. Specifically, it is shown how a Landmark routing solution could be implemented within RINA. Finally, efforts are also devoted to analyse the importance of path selection for ensuring QoS requirements and how it is not required to reach extremes solutions, like the use of connections, to provide the required services.
