Abstract

[ANGLÈS]Lately, the emerging paradigm of Software-Defined Networking has grown in presence and claims to simplify future networking. The decoupling of network control and forwarding plane proposed by this architecture allows the control of the entire network behavior by means of a logically centralized software program (controller). Such separation of planes opens the way to Network Virtualization, which provides users a logical abstraction of underlying network resources. However, network virtualization requires a mapping phase of the virtual resources over the physical resources, which is not trivial, formally known as the virtual network embedding problem. The present document focuses in the development of a variant of one of the proposed strategies to solve this critical step, prioritizing the real-time response. The proposed algorithm uses the properties offered by the Paths Algebra mathematical framework to provide a flexible environment where it is possible to combine any number of both linear and non-linear metrics. In addition, it is also used the multi-lexical ordination, a criterion to better distinguish paths that may be considered equal by other approaches. Such algorithm has been implemented as a software application that provides a simulation environment where the virtual network embedding process can be tested for any given topology. Subsequently, all the algorithm features have been checked in a set of performance tests, focusing on those oriented to the commitment among the real-time response and the quality of the embedding solutions. In general, testing results are very promising even in densely populated backbone topologies, where the number of alternative paths among each possible pair of origin and destination nodes grows exponentially. [CASTELLÀ] Software-Defined Networking (o Redes Definidas por Software) es un nuevo paradigma que tiene como objetivo simplificar la creación y gestión de redes de ordenadores. El desacoplamiento entre el control de la red y el plano de reenvío propuesto por esta arquitectura permite el control de todo el comportamiento de la red mediante un elemento lógico centralizado, llamado controlador. Esta separación de los planos abre la puerta a la virtualización de redes, proporcionando a los usuarios una abstracción lógica de los recursos de red subyacentes. Sin embargo, la virtualización de red requiere de una fase de asignación de los recursos virtuales a los recursos físicos, que no es trivial y que se conoce formalmente como el problema de incrustación de redes virtuales. El presente documento se centra en el desarrollo de una variante de una de las estrategias propuestas para resolver este paso crítico, dando prioridad a la respuesta en tiempo real. El algoritmo propuesto utiliza las propiedades ofrecidas por el marco matemático de Paths Algebra (o Álgebra de Caminos) para proporcionar un entorno flexible donde es posible combinar cualquier número de métricas lineales y no lineales. Además, también utiliza la ordenación multi-léxica, un criterio para distinguir mejor aquellos caminos que podrían ser considerados equivalentes por otros enfoques. Este algoritmo se ha implementado como una aplicación de software que proporciona un entorno de simulación en el que se puede probar el proceso de incrustación de redes virtuales para cualquier topología. Posteriormente, se han comprobado todas las características del algoritmo mediante un conjunto de pruebas de rendimiento, priorizando aquellas orientadas al compromiso entre la respuesta en tiempo real y la calidad de las soluciones de incrustación. En general, los resultados de las pruebas son muy prometedores incluso en topologías de redes troncales densamente pobladas, donde el número de caminos alternativos entre cada posible nodo origen y destino crece exponencialmente.

Abstract

International audience; In Software-Defined Networking, near-real-time collection of flow-level statistics provided by OpenFlow (e.g. byte count) is needed for control and management applications like traffic engineering, heavy hitters detection, attack detection, etc. The practical way to do this near-real-time collection is a periodic collection at high frequency. However, periodic polling may generate a lot of overheads expressed by the number of OpenFlow request and reply messages on the control network. To handle these overheads, adaptive techniques based on the pull model were proposed. But we can do better by detaching from the classical OpenFlow request-reply model for the particular case of periodic statistics collection. In light of this, we propose a push and prediction based adaptive collection to handle efficiently periodic OpenFlow statistics collection while maintaining good accuracy. We utilize the Ryu Controller and Mininet to implement our solution and then we carry out intensive experiments using real-world traces. The results show that our proposed approach can reduce the number of pushed messages up to 75% compared to a fixed periodic collection with a very good accuracy represented by a collection error of less than 0.5%.