To reduce the latency of applications and to cope with an increasing number of networked devices, fog computing places processing and storage resources towards the edge of the networks. This leads to the creation of small data centers, with heterogeneous hardware, and communication protocols. Despite the advantages of this approach, many concerns arise regarding the share of the energy consumed by network infrastructures and their applications on the world’s GHG emissions. To estimate this impact, existing works propose to make use of testbeds or models to better understand the cost of different parts of the network. However, existing works have to balance between accurate results for small-scale measurements and large-scale experiments with coarse-grained results. The goal of this thesis is to further study the tradeoff between scalability and accuracy to measure the emissions of end-to-end fog infrastructures at scale. We propose models based on flow-level simulation to evaluate the performance and the energy consumed by Wi-Fi networks and microservice applications. Furthermore, we combine our contributions with previous flow-level models to study a realistic fog infrastructure at scale. The models proposed in this thesis are available within the open-source SimGrid simulation framework, with reproducible validation experiments.
- Pierre Sens : LIP6
- Adrian Friday Université de Lancaster
- François Taïani : Irisa
- Martin Quinson : ENS Rennes
- Anne-Cécile Orgerie : CNRS
L’accès ne sera pas autorisé sans inscription préalable. Par ailleurs, les visiteurs ne porteront ni bagage ni sac.