OBJECTIVES OF THE COURSE:
In a first step, we propose a brief state of the art on CPU and hardware architectures. The performance criteria of a blade server (the basic computing unit within a datacenter) are analyzed (CPU speed, multithread vs. multicore, energy consumption etc.). We describe then into details the rationale of the typical 2D-fat-tree optical backplane network architecture of a datacenter. The pros and cons of this architecture are discussed in reference to the potentialities of all-optical technologies. The basic rationale of virtual machines (VM) assignment to physical machines (PM) and migration strategies is presented via bin packing theory. After these preliminary considerations, the evolution from hypervisors to Virtual Containers (VC) such as Kubernetes and Docker is briefly discussed. An original focus is dedicated to the impact of heat dissipation onto real-time VM migrations. In this perspective, we show how a Pareto optimization approach can be considered. We conclude this first section by a brief presentation of Cloud standardization bodies and the three Cloud business models considered up to now.
The second part of this course deals with the most innovative aspect of the Cloud domain, namely Cloud Radio Access Networks (CRAN) and Mobile Edge Computing (MEC). CRAN refers to the innovative access infrastructure that will enable to deport Cloud intelligence close to fix/mobile end-users, including IoT sensors/actuators. Autonomous vehicles (AV) and more extensively, Intelligent Transportation Systems (ITS), the first two pillar applications of 5G, rely on Cloud and IoT. In this perspective, we depict into details the basic principles of Digital Radio-over-Fiber (D-RoF) and the two underlying standards, namely CPRI and OBSAI. In our conclusion, we depict the open perspectives offered by 5G/MEC for future intelligent transportation systems (ITS).
