多vCPU环境中基于容器的科学工作流调度策略

现有科学工作流调度研究较少考虑计算资源的多道程序设计，难以同时实现有效的容器共享并优化任务并行度与资源利用率。为了解决以上难点，文章提出了一种分布式多vCPU环境中基于容器技术的分段式工作流调度策略。该策略通过分段调度方法，降低启发式算法的解空间大小，使用带遗传算子的自适应离散粒子群优化算法(ADPSOGA),在设备使用成本的约束下优化各个工作流的完成时间，并制定一种容器与设备间的动态伸缩方案，实现容器的复用并解决单个设备中任务并行时的资源争用问题。结果表明：ADPSOGA的性能优于其他同类启发式算法，并且分段调度方法与容器伸缩方案在工作流调度方面表现出良好的性能，能很好地适应因任务并行度增加所带来的影响。

Container-driven Scheduling Strategy for Scientific Workflows in Multi-vCPU Environments

Existing scientific workflow scheduling studies barely consider the multi-channel programming of computational resources, which makes it hard to simultaneously achieve effective container sharing and optimize task parallelism and resource utilization. In order to solve the above problems, a segmented workflow scheduling strategy based on container technology in multi-vCPU environment is proposed. It reduces the solution space size of the heuristic algorithm through a segmented scheduling approach. And it uses an adaptive discrete particle swarm optimization algorithm with genetic operators (ADPSOGA) to optimize the completion time of each workflow under the constraint of device rental cost. In addition, a dynamic scaling scheme between containers and devices is proposed to reuse containers and solve the problems related to resource contention when tasks are parallel in a device. The results indicate that ADPSOGA outperforms other similar heuristics algorithms, and the segmented scheduling approach with container scaling scheme shows good performance in workflow scheduling and can adapt well to the impact of increasing task parallelism.