Improving the performance of Virtualized Network Services based on NFV and SDN

Network Functions Virtualisation (NFV) proposes to move all the traditional network appliances, which require dedicated physical machine, onto virtualised environment (e.g,. Virtual Machine). In this way, many of the current physical devices present in the infrastructure are replaced with standard high volume servers, which could be located in Datacenters, at the edge of the network and in the end user premises. This enables a reduction of the required physical resources thanks to the use of virtualization technologies, already used in cloud computing, and allows services to be more dynamic and scalable. However, differently from traditional cloud applications which are rather demanding in terms of CPU power, network applications are mostly I/O bound, hence the virtualization technologies in use (either standard VM-based or lightweight ones) need to be improved to maximize the network performance. A series of Virtual Network Functions (VNFs) can be connected to each other thanks to Software-Defined Networks (SDN) technologies (e.g., OpenFlow) to create a Network Function Forwarding Graph (NF-FG) that processes the network traffic in the configured order of the graph. Using NF-FGs it is possible to create arbitrary chains of services, and transparently configure different virtualized network services, which can be dynamically instantiated and rearranges depending on the requested service and its requirements. However, the above virtualized technologies are rather demanding in terms of hardware resources (mainly CPU and memory), which may have a non-negligible impact on the cost of providing the services according to this paradigm. This thesis will investigate this problem, proposing a set of solutions that enable the novel NFV paradigm to be efficiently used, hence being able to guarantee both flexibility and efficiency in future network services.