Context. Edge Computing (EC) represents one of the major architectures’ transformations
that will enable future 5G/6G networks to support a broad set of new services’ paradigms. In
Multi-access EC (MEC), computational resources (RAM, CPU, GPU) are deployed in nodes
placed at the edge of the network, close to the devices, i.e., close to base stations, central
offices, access points, or at the border of routes in the form of Road Side Units (RSUs). This
enables services with tight latency constraints, e.g., augmented reality or autonomous driving,
which cannot be currently offered under the legacy paradigm of Cloud Computing. The
distribution of applications with optimal location of application’ components and data may lead
to significant reductions in energy consumption.
Problem. Edge nodes are usually assumed to be mono-application. An application provider,
e.g., a car manufacturer, would deploy its own road side units running autonomous driving
functions. Another application provider, e.g., augmented reality, would separately deploy its
own edge nodes close to the potential users, etc. Such a deployment could lead to a
proliferation of multiple computing nodes, each turned on ready to satisfy user’s request, with
consequent explosion of energy consumption.
Key idea. In our previous work [1] we have proposed “multi-tenant edge computing”, where
multiple application providers form a coalition to co-invest in the deployment of edge
infrastructure and we have studied via coalitional game theory how to share the cost of and
the benefits from such deployment. We believe multi-tenant edge computing is an opportunity
for reducing the impact of EC: by forming coalitions, application providers can concentrate
traffic in fewer resources. The first gain is that fewer nodes up and running translate to lesser
energy consumption. Second, the geographical diversity of edge nodes allows to make optimal
choices of the energy source. Indeed, in a coalition formed by multiple edge infrastructure
owners and multiple application providers, the former can deploy nodes in different
geographical regions and the latter can redirect computation with relatively large delay
tolerance to the nodes with greener energy sources, i.e., nodes whose energy mix is rich in
renewable sources, e.g., solar or wind.
Activities. We will develop a model based on coalitional game theory, adapting results that
we obtained in the smart grids domain for the sharing of local renewable energy production
and of storage (see [2] to [10]), a problem that has several similarities with ours and for which
we proposed advanced solutions that received top level international awards.
In our model, players are multiple edge infrastructure owners and application providers. Such
players are interested in co-investing in the deployment of a geographically distributed set of
nodes, thus sharing the capital and operational cost and the benefits. We will consider users
generating a mix of latency-sensitive and best-effort traffic, which will optimally be routed to
the appropriate edge nodes, considering the trade-off between latency and energy efficiency.
We will study under which conditions selfish behavior of such players leads to energy
efficiency. In this sense, we will study the impact of the price of energy, which in our model is
an exogenous parameter.
Activity organization. The problem will be studied by an intern for 6 to 12 months, co-
supervised at Télécom Paris and Télécom SudParis by the proponents of this project.