Edge computing is a promising paradigm that relies on heterogeneous computing resources located at the edge of the network, close to the end-user. Hence, applications with latency-sensitive and compute-intensive tasks rely on edge resources to offload and complete such tasks. In order to support non-intermittent service in case of user mobility, most of existing approaches focus on how to accelerate the handoff transfer time and not how to reduce its frequency. Moreover, the handoff mechanisms used in cellular networks do not consider the computational workload and therefore are not directly applicable to the edge offloading scenario. Considering dense edge deployment, it is vital to select the optimal edge node for offloading. Therefore, we take into consideration bandwidth, processor speed and latency capabilities in the proposed fuzzy logic node selection algorithm. We evaluate the improvements of the proposed selection, for perceived response time objective, in comparison to offloading to the closest or highest bandwidth node. In addition, we propose a handoff controller, to meet the same performance objective, when the user is moving further from the currently selected edge node. We evaluate our approach by offloading Directed Acyclic Graph (DAG) models of real-world mobile applications. The results show that we can significantly reduce both application response time and monetary cost of execution, by controlling the number of handoffs among edge nodes.

Top