A Bilevel Decomposition Approach for Many Homogeneous Computing Tasks Scheduling in Software-Defined Industrial Networks

To confront the great challenge of industrial big data, the software-defined industrial networks (SDINs) are introduced to dynamically coordinate these data flows among the heterogeneous and distributed computing resources. Deciding how to more efficiently schedule many homogeneous computing tasks, which extensively appear in SDIN, becomes of critical importance. To this end, this article first illustrates some related notations and assumptions of the homogeneous computing tasks and computing networks, from which a new targeted optimization model is formulated. Then, the model is significantly enhanced by reformulating all the nonlinear constraints and inventively establishing the symmetry-breaking constraints and computation time cuts. Furthermore, considering the computational complexity, the scheduling process with many homogeneous computing tasks is further viewed as two associated phases: computing nodes assignment and tasks sequencing. As a result, a novel bilevel decomposition algorithm is proposed using Lagrangian decomposition and a new form of Lagrangian relaxation. Finally, a real industrial scenario is chosen, and three comparison algorithms are used to demonstrate the preponderant performance of the proposed algorithm.

Automated summary

To tackle the challenge of industrial big data, software-defined industrial networks are used to manage data flows among heterogeneous and distributed computing resources. This article proposes a novel optimization model for efficiently scheduling homogeneous computing tasks in these networks.