Towards Efficient Allocation of Tasks in Dag-Based Workflows Across Federated Fog Systems

DAG-based (Directed Acyclic Graph) applications, deployed on fog computing platforms (mini-data centers), are at the core of modern industrial systems-especially those in remote locations, which are often subjected to harsh, disaster-prone conditions and frequently face unreliable or no cloud connectivity. When disasters occur and multiple time-sensitive DAGs need to be processed, the inherent inelasticity of local fog systems arises as a barrier to meeting latency constraints. To achieve cloud-like elasticity in such resource-limited environments, we propose resource allocation strategies based on Mixed-Integer Linear Programming (MILP) to optimally “partition” the DAG and “approximately process” its tasks in a federated fog system. To make our approximate computing strategy feasible for largescale DAGs, we developed a second method based on the greedy approach. Our experimental results demonstrate that our strategies establish elasticity across federated fog systems, as well as minimize the approximation level while meeting the deadline constraints of the DAGs.