Abstract: For the existing task offloading and scheduling scheme, focus on the problem of computing process is complex, easily overlook the multiple task properties and the parallelism of subtasks, et al. this paper proposes a fine-grained task scheduling strategy under the cooperation of cloud, edge and terminal. Firstly, a hierarchical network architecture with cloud, edge and terminal layers and introduce SDN (software defined network) is built to achieve efficient network information management and balance the edge load. Then the task scheduling strategy is proposed with two task serving patterns and three execution mechanisms: for tasks with pattern of “served as a whole”, an optimization problem is built to minimize delay and energy consumption jointly, thus achieving the optimal execution mechanisms; for tasks with pattern of “served with segmentations”, a SSPM (subtask scheduling parallelism maximization) algorithm based on dynamic list scheduling is designed to maximize the parallelism degree of subtasks scheduling as much as possible. Finally, the simulation is carried out on NS-3 platform, and results show that the scheme outperforms others.

Key words: cloud-edge-terminal cooperation, task scheduling, task serving pattern, task execution mechanism

摘要： 针对已有的任务卸载及调度方案大多计算过程复杂、易忽略任务多属性及子任务间可并行性等问题，设计了一种云-边-端协作网络中的细粒度任务调度策略。构建了一个涵盖云-边-端三层的网络架构并引入SDN（software defined network）来实现高效的网络信息管理和边缘负载均衡。提出了包含两种任务服务模式（整体服务模式、分割服务模式）及三种执行机制（云执行、边执行、端执行）的任务调度策略：对于整体服务模式下的任务，设计了一个时延与能耗联合优化的优化问题来获取对应任务的最佳执行机制；对于分割服务模式下的任务，利用改进的动态列表调度方法设计了一种子任务调度并行度最大化（subtask scheduling parallelism maximization, SSPM）算法来最大化分割后被调度子任务的并行度。在NS-3平台上进行仿真实验，仿真结果表明该策略的性能表现优于其他对比方案。

关键词: 云-边-端协作, 任务调度, 任务服务模式, 任务执行机制