Optimal Placement of Delay-constrained In-Network Computing Tasks at the Edge with Minimum Data Exchange

Task offloading to the edge is gaining momentum to support compute-intensive interactive applications which, on the one hand, can hardly run on resource-limited consumer devices and, on the other hand, may suffer from running in the cloud due to the strict delay constraints. The availability of network nodes with heterogeneous capabilities in the distributed edge infrastructure paves the way to groundbreaking in-network computing scenarios, but makes the computing task allocation decision more challenging. The straightforward approach of offloading the computation task to the edge node that is the nearest to the data source may lead to performance inefficiencies. Indeed, such edge node may easily get overloaded, thus failing to ensure low-latency task execution. A more judicious strategy is required which accounts for the edge nodes' processing capabilities and for the queuing delay accumulated when tasks wait before being executed as well as for their connectivity. In this paper, we propose a novel optimal computing task allocation strategy aimed at minimizing the network resources usage, while bounding the execution latency at the edge node acting as the task executor. We formulate the optimal task allocation through an integer linear programming problem, assuming an edge infrastructure managed through software-defined networking. Achieved results show that the proposal meets the targeted objectives under all the considered simulation settings and significantly outperforms other benchmark solutions.