MSc Graduation project performance and resource optimization of virtualized data center architectures
Driven by the cloud computing paradigm and Internet applications, data centres (DC) experience a steady annual increase of over 70% in the amount of traffic. This huge increase of traffic requires not only architectural and technological changes in the DC networks (DCN) to enhance higher capacity and low latency to support the demanding traffic volume, but also an efficient software define networking (SDN) control plane for optimal resources utilization through virtualization technologies to significantly improve the efficiency, resiliency, responsiveness, and greenness of DCs. Recently, a novel DCN architecture, OPSquare, based on fast optical switches (FOS) that potentially meets the requirements of high bandwidth and low latency has been proposed (see Fig. 1). Despite preliminary experimental and numerical demonstration, no investigation on the potential performance enhancement that the SDN virtualization can offer by applying novel load balancing, network slicing and resource optimization algorithms to the OPSquare DCN architecture.
The project objective is to investigate, employing the real-time Mininet emulator, the OPSquare performance and the resource optimization exploiting SDN virtualization, novel load balancing, network slicing, and resource optimization algorithms. The assignment comprises the following main tasks:
- To develop the SDN controller for the OPSquare architecture in the Mininet emulator using Python. As technical skills, the candidate will master Mininet (real time network emulator), Python program skill for implementing SDN APIs, and the knowledge of the OpenFlow protocol for SDN control plane
-To develop the SDN interfaces of the optical switches and the Top-of-the-Rack (TOR) switches of the OPSquare architecture also in Mininet
.- To generate multiple virtual slices and analyze the OPSquare DCN performance in terms of packet loss, latency, throughput, and resource allocation
.- To implement and compare the static and dynamic load balancing algorithms, and to study the optimized algorithm for OPSquare
- To study the networking slicing including physical and logical scenarios shown in Fig.1