In this paper, we consider the computation offloading optimization problem with heterogeneous resources in a fog-cloud computing system. The problem is common in many real-time and mobile applications, where tasks are massive and computation-intensive, and the computing resources could involve both fog devices and cloud platforms. The challenges lie in proposing effective, efficient and robust algorithms with the objectives of minimizing both the total delay and the energy consumption. A bi-objective task scheduling model is formulated, in which the queuing models for the delay estimation and the energy consumption models for heterogeneous resources are introduced. A Pareto-optimization-based Massive Task Scheduling Framework is proposed to schedule massive tasks within one time unit. It starts from a non-dominated solution set obtained by the energy-ware and the transmission delay-aware local search procedures. A tree-based local search method is proposed to further improve the non-dominated solutions. The proposed algorithm is compared to four classical algorithms for the similar problems. Their performances are evaluated by the Pareto-optimization metrics on multiple aspects. Experimental results demonstrate the effectiveness and robustness of the proposal for the problem under study.

在本文中，我们考虑了雾云计算系统中异构资源的计算卸载优化问题。这个问题在许多实时和移动应用程序中很常见，在这些应用程序中，任务非常繁重且计算量很大，并且计算资源可能涉及雾化设备和云平台。挑战在于提出有效，高效和鲁棒的算法，以最小化总延迟和能耗为目标。建立了双目标任务调度模型，引入了时延估计的排队模型和异构资源的能耗模型。提出了一种基于帕累托优化的大规模任务调度框架，用于在一个时间单位内调度大规模任务。它从能源软件和感知传输延迟的本地搜索程序获得的非支配解决方案集开始。提出了一种基于树的局部搜索方法，以进一步改善非支配解。针对相似的问题，将提出的算法与四种经典算法进行了比较。它们的性能由帕累托优化指标在多个方面进行评估。实验结果证明了针对所研究问题的提议的有效性和鲁棒性。