Towards user-centric, switching cost-aware fog node selection strategies

• Published in: Future generation computer systems Vol. 117; pp. 359 - 368
• Main Authors: Rejiba, Zeineb, Masip-Bruin, Xavier, Marín-Tordera, Eva
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.04.2021
• Subjects: Edge computing; Fog computing; Mobility management; Multi-armed bandits; Switching cost; Fog computing; Multi-armed bandits; Switching cost; Mobility management; Edge computing
• ISSN: 0167-739X; 1872-7115
• DOI: 10.1016/j.future.2020.12.006

In order to address high latency issues that may arise when executing time-critical applications at the cloud side, the novel fog computing paradigm has emerged, thus enabling the execution of such applications within computation nodes present at the edge of the network. While executing such applications, a user may be moving in an area where a high number of heterogeneous fog nodes (FNs) co-exist. This makes the problem of selecting the most appropriate fog node to execute the user’s tasks challenging, especially since the set of visible FNs dynamically changes. Therefore, to deal with the uncertain and dynamic nature of such a fog computing environment, we model the FN selection problem using multi-armed bandits. However, standard solutions for the bandit problem are not tailored for scenarios with changing FN availabilities. In addition, since switching from one FN to the other causes a switching cost, such solutions lead to accumulating a high switching cost. Therefore, to address these issues, we first propose a block-based FN selection scheme, where switching among FNs is not allowed during a block of timeslots. We also propose a greedy approach, where FNs having a sufficiently good delay performance are selected in a greedy manner. Simulation results reveal that both approaches significantly improve the FN selection performance. In particular, we found that the block-based selection results in the lowest switching costs, whereas the greedy selection achieves the best overall performance. •User mobility in a high density fog computing environment makes the task of fog node selection particularly challenging.•Switching from one fog node to the other should be carefully controlled in order to keep the overall task execution delays low.•A fog node selection strategy using block-based multi-armed bandits can reduce the accumulated switching costs.•An adaptive greedy FN selection approach significantly reduces the average cumulative delay for a user’s tasks.