Resumen de Mobility-aware mechanisms for fog node discovery and selection
The recent development of delay-sensitive applications has led to the emergence of the fog computing paradigm. Within this paradigm, computation nodes present at the edge of the network can act as fog nodes (FNs) capable of processing users' tasks, thus resulting in latency reductions compared to the existing cloud-based execution model.
In order to realize the full potential of fog computing, new research questions have arised, mainly due to the dynamic and heterogeneous fog computing context. This thesis focuses on the following questions in particular: How can a user detect the presence of a nearby FN? How should a user on the move adapt its FN discovery strategy, according to its changing context? How should an FN be selected , in the case of user mobility and FN mobility? These questions will be addressed throughout the different contributions of this thesis.
The first contribution consists in proposing a discovery solution allowing a user to become aware of the existence of a nearby FN. Using our solution, the FN advertizes its presence using custom WiFi beacons, which will be detected by the user via a scan process. An implementation of this approach has been developed and its evaluation results have shown that it results in a non-negligible energy consumption given its use of WiFi.
This has led to our second contribution, which aims at improving the WiFi scan performed in our discovery approach, especially in the case of user mobility. At a first stage, this improvement consisted in embedding information about the topology of the FNs in the beacons the user receives from previous FNs. We have shown that by adapting the scan behavior based on this information, considerable energy savings can be achieved, while guaranteeing a high discovery rate. However, as this approach is associated with a restrictive FN topology structure, we proposed a different alternative, at a second stage. This alternative leverages the history of cellular context information as an indicator allowing the user to infer whether an FN may be present in its current location. If so, the scan will be enabled. Otherwise, it is disabled. The simulation results comparing different classification algorithms have shown that a sequence-based model, such as a hidden-Markov model is able to effectively predict the FN presence in the current user location.
While the previous approaches have focused on a sparse FN deployment, our third contribution considers a high density of FNs. Consequently, as there are multiple nearby FNs that can process the user's tasks, it is important to derive a suitable FN selection strategy. This strategy should consider the time-varying set of FNs caused by the user's mobility. Besides, it should minimize the number of switches from one FN to another, in order to maintain a good quality of service. With these considerations in mind, we have shown that an adaptive greedy approach, that selects an FN having a good-enough delay estimate, achieves the best results.
Finally, unlike the previous contribution, where the focus has been on FN selection when the user is mobile, our final contribution deals with mobile vehicular FNs (VFNs). Given the mobility of such VFNs, it is important to make the most of their resources, since they are only available for a short time at a given area. So, we propose that, in order to select an appropriate VFN for a given task, a reference roadside unit (RSU) responsible for task assignment can use advice from a neighbor RSU. This advice consists in the VFN that will result in the lowest delay for the current task, based on the experience of the neighbor RSU. The results have shown that, using the provided advice, the reference RSU can observe significant delay reductions.
All in all, the proposed contributions have addressed various problems that may arise in a fog computing context and the obtained results can be used to guide the development of the building blocks of future fog computing solutions.
