A Large-Scale Wired Network Energy Model for Flow-Level Simulations

Abstract

The use of simulators to predict network energy consumption is a good way for scientists to improve and develop new algorithms and also to assess them. However, the average size of a network platforms is continuously increasing with the emergence of new technologies like the Internet Of Things and Fog Computing. Packet-level simulators start to reach their limits in terms of performance and this calls for newer solutions in the domain of large-scale platform energy models. In this paper, we propose two energy models for wired networks adapted to flow level simulators in order to estimate the energy consumption of large platforms. An evaluation of these models is proposed and it demonstrates their applicability in practice and their accuracy. Indeed, we obtain simulation results with a relative error lower than 4% compared to an ns-3-based solution, and our flow-based simulation is 120 times faster.

Co-simulation of an electrical distribution network and its supervision communication network

“…The drawback of SimGrid compared to packet-level simulators like NS-3 is that currently, its model is limited to TCP/IP and does not include wireless communications. Yet, if other communication protocols are required, we can use the ad-hoc coupling between SimGrid and NS-3 that has been already developed [13]. Thanks to this coupling, SimGrid can use any NS-3 model instead of its own flow-based model.…”

“…However, these tools do not scale up well because they simulate each packet transfer on the network. Then, it may take orders of magnitude longer than the simulated time to simulate large and complex communication network topology [13]. Hence, packet-level simulators do not seem adapted to conduct experiments on energy management in smart grids that may consider timescales ranging from seconds to several years.…”

“…The SimGrid platform [4] is dedicated to the simulation of large-scale distributed systems. It embeds a validated flowlevel TCP/IP model that is faster by orders of magnitude than packet-level approaches [13]. Moreover, it has been extensively demonstrated that this model is also reasonably accurate [29].…”

Optimizing Green Energy Consumption of Fog Computing Architectures

“…The power model used for the network consumption comes from Guegan and al. [13]. The energy consumption of a link corresponds to the consumption of the its two Ethernet devices' ports connected by the link.…”

“…5) Network Power Model: As explained in Section III-B, the network power model is from a paper by Guegan and al. [13]. Values used are described in Table III.…”