Matthias Frei

In the future, data exchange will no longer take place exclusively between the cloud (or a server in a data center) and a mobile device. Instead, communication between devices will be established directly on the basis of application relationships in order to realize immersive applications, automated driving or virtual reality. To this end, 5G and future network technologies are increasingly following the data-centric paradigm in their design, in which, among other things, the increasing relevance of direct device communication is taken into account. Another elementary development also contributes to this: Computing or information resources are no longer provided exclusively by cloud servers.
Multi-Access Edge Computing (MEC) is part of current research and deals with the provision of resources on distributed edge nodes. For example, MEC instances can be located close to base stations to serve applications with special requirements, such as low latency, low-variance jitter, high bandwidth, or privacy requirements close to the end device. Over time, services will emerge whose components can be deployed literally anywhere and in a distributed manner - without the need to consider a mandatory hierarchical network topology. In addition to a cloud instance, a service can therefore also be operated on the edge instance in the vicinity, e.g. a mobile radio base station, a traffic control system or even a neighboring user equipment (UE). Multi-level MEC constellations are also possible. A homogeneous technology stack that extends cloud computing enables a data-centric architecture that can simultaneously accommodate stringent service requirements.
The resulting architecture can be viewed from two perspectives. In terms of network communication, MEC resources are accessible via only a few links or hops. This geographical or topological proximity means that the links are not overloaded, which results in the aforementioned performance advantages. With regard to the services provided, a MEC orchestrator can dynamically adapt service deployments on compute nodes to the current situation and integrate resources into the topology or remove them, for example, to save energy. In addition to orchestration decisions, the movement of nodes also leads to a change in the network topology. In order to exploit the full potential of MEC and thus also to be able to operate services that rely on MEC resources, both perspectives must be combined in a meaningful way.
In static environments, MEC resources can usually be planned well in advance. It becomes a challenge especially when the mentioned dynamic topology changes or mobility of UEs affect the overall system. One of the key questions that arises is: Can the communication requirements of MEC-dependent services, which are necessary for the smooth implementation of the service, be met at all times?
The research project deals with the selection of the best MEC resources, for example from a UE point of view, as well as the, from a network point of view, best locations for orchestrators to provide the services. In particular, the focus is on the current network and topology situation in combination with the strict communication requirements of services that need MEC resources. Strategies and algorithms, for example based on graphs, are developed, implemented and evaluated. Verification takes place through system-level simulations and real-world deployment.