Highly scalable resource supply capacity of cloud computing has greatly improved the execution speed of workflow applications, however, traditional workflow scheduling algorithms which focus on the optimization of makespan (execution time) of workflows, become inappropriate for the design of large-scale workflow systems. Workflow scheduling in cloud computing is particularly a multiobjective optimization problem, in which many critical issues besides the execution time of workflows should be taken into account. Although many heuristics and meta-heuristics have been proposed to solve this problem, most of them cannot produce satisfactory cost-makespan tradeoffs and have a long time overhead. In this paper, we propose an efficient heuristic named CMSWC (Cost and Makespan Scheduling of Workflows in the Cloud) to solve the workflow scheduling problem, by simultaneously minimizing cost and makespan of workflows. CMSCW follows a two-phase list scheduling philosophy: ranking and mapping. Furthermore, CMSCW incorporates with three designs specifically for the multiobjective challenges: (i) The mapping phase is designed to avoid exploring useless resources for tasks, which significantly narrows down the search space. (ii) A new method is proposed to select non-dominated solutions, by combining the quick non-dominated sorting approach and Shift-Based Density Estimation (SDE) based crowding distance. (iii) Several elitist study strategies are designed to make solutions close to the true Pareto front as well as avoid trapping into local optimum. Extensive experiments on real-life workflows demonstrate that our approach can generate better cost-makespan tradeoff fronts than that of several state-of-the-art approaches.

云计算的高度可扩展的资源供应能力极大地提高了工作流应用程序的执行速度，但是，传统的工作流调度算法侧重于工作流的有效期（执行时间）的优化，因此不适用于大规模工作流系统的设计。云计算中的工作流调度尤其是一个多目标优化问题，除工作流的执行时间外，还应考虑许多关键问题。尽管已经提出了许多启发式方法和元启发式方法来解决该问题，但是它们中的大多数不能产生令人满意的成本-取舍权衡并且具有长的时间开销。在本文中，我们提出了一种有效的启发式方法CMSWC（Cost和Makespan计划在云中的工作流）来解决工作流计划问题，同时降低成本和缩短工作流程的时间。CMSCW遵循两阶段的列表调度原理：排名和映射。此外，CMSCW结合了三种专门针对多目标挑战的设计：（i）映射阶段的设计避免了为任务探索无用的资源，从而极大地缩小了搜索空间。（ii）通过结合快速非支配排序方法和基于位移的密度估计（SDE）拥挤距离，提出了一种选择非支配解决方案的新方法。（iii）设计了几种精英研究策略，以使解决方案接近真正的帕累托前沿，并避免陷入局部最优状态。