The hybrid cloud is a crucial solution to overcome the limited resources of the private cloud and efficiently execute large-scale workflow due to its easy scalability and ability to guarantee data privacy. However, most of the existing studies on multi-objective workflow scheduling in a hybrid cloud treat the problem as a black box and perform overall optimization of large-scale decision variables, which will lead to the inefficient search of the solution space. In order to compensate for the above shortcomings, this paper proposes a workflow dynamic neighborhood search (WDNS) scheduling algorithm to simultaneously optimize the makespan, cost, and energy consumption of workflow execution in a hybrid cloud. Firstly, based on the data dependencies among workflow tasks, a dynamic neighborhood grouping strategy is proposed to divide decision variables with strong dependencies into the same group, thus effectively increasing the possibility of simultaneously optimizing interdependent variables. Then, based on the grouping strategy, new crossover and mutation operators are designed to search for feasible solutions, aiming to take advantage of divide-and-conquer to improve the search efficiency. Finally, in the context of 20 real-world workflows, our proposed WDNS algorithm was compared with four state-of-the-art algorithms. The comparison results confirm that WDNS outperformed the four algorithms across all 20 test cases in both hypervolume and inverted generational distance metrics.

中文翻译：

---

基于动态邻域分组的混合云工作流多目标调度算法


混合云是克服私有云有限资源并有效执行大规模工作流程的关键解决方案，因为它具有简单的可扩展性和保证数据隐私的能力。然而，现有的关于混合云中多目标工作流调度的研究大多将问题视为一个黑匣子，对大规模决策变量进行整体优化，这将导致对解决方案空间的低效搜索。为了弥补上述不足，本文提出了一种工作流动态邻域搜索 （WDNS） 调度算法，以同时优化混合云中工作流执行的 makespan、成本和能耗。首先，基于工作流任务之间的数据依赖关系，提出一种动态邻域分组策略，将具有强依赖关系的决策变量划分为同一组，从而有效增加了同时优化互相关变量的可能性。然后，基于分组策略，设计新的交叉和突变算子来寻找可行的解，旨在利用分而治之来提高搜索效率。最后，在 20 个实际工作流程的背景下，我们提出的 WDNS 算法与 4 种最先进的算法进行了比较。比较结果证实，WDNS 在所有 20 个测试用例中，在超容量和倒置代际距离指标上都优于四种算法。