协作式边缘计算中网络感知的任务调度技术研究

协作式边缘计算作为一种新型的计算模型，通过分布在不同地理位置的边缘资源的相互协作完成特定的计算任务，并广泛应用在物联网、智能交通、智慧城市等领域。任务调度是保障任务协作执行性能的重要手段。已有任务调度方法关注对计算任务本身的调度，忽略了任务之间的数据依赖引发的网络传输和调度，因此在资源高度分散和受限的协作式边缘计算中，无法获取较好的性能。本项目将针对协作式边缘计算中不同的协作类型，提出网络感知的任务调度技术，即对计算任务以及网络层的数据传输进行联合优化和调度，旨在提高应用系统的性能。具体研究内容包括：（1）垂直协作下针对工作流应用的任务卸载和协同在线调度，以及针对数据流应用的任务卸载与资源分配；（2）水平协作下联合数据缓存与传输的任务分配，以及保障QoS的大规模并发任务在线迁移。预期研究成果有望为突破分布式网络系统中应用性能瓶颈提供新思路和新方法。

As a new computing paradigm, Collaborative Edge Computing (CEC) relies on the collaboration among geographically distributed resources to complete specific tasks. It has been widely used in many application domains such as IoT, intelligent transportation systems, smart cities and so on. Task scheduling is an important technique to guarantee the performance of the task execution in CEC. Existing task scheduling methods focus on the allocation of computational tasks onto a set of resources, while they neglect the scheduling of network layer transmissions arising from the data dependency among the tasks. Thus, these methods can not achieve satisfactory performance as they are applied in CEC where the resources are highly distributed at various locations and constrained. In this project, we focus on two types of collaboration in CEC, and propose network aware task scheduling and resource allocation, which jointly schedules the computational tasks and the arised data transmissions in the network, aiming at optimizing the task execution performance. In specific, our research will include two folds.1) In terms of vertical collaboration, we will study task offloading and cooperative online scheduling for workflow applications, and task offloading and resource allocation for data stream applications. 2) In terms of horizontal collaboration, we will study the joint optimization of data caching, data transmission and task allocation. Furthermore, we will study new methods for large scale and concurrent QoS guaranteed live task migrations. Our proposed research provides new methodologies and insights for breaking the performance bottleneck in distributed networking systems.