面向多边缘云的资源调度与协作技术研究

To cope with the explosive growth of edge devices and data, edge computing emerged as a promising way by performing data processing at the edge of the network, near the source of the data. While more and more data processing jobs require resources from multiple edge clouds, inappropriate or inefficient resource allocation for such jobs probably impairs quality-of-service. Therefore, we should systematically design a distributed resource scheduling framework for heterogeneous edge clouds. Our preliminary case studies reveal that resource demands of cross-edge jobs expose strong patterns in both time and space. These findings motivate us to exploit pattern-driven approaches that guide resource pre-allocation. We will also design a topology-aware multi-dimensional framework for estimating edge capacity, make use of the collaboration among users, edge providers, and service providers to optimize resource scheduling, and protect key jobs through intelligent selective replication of data and tasks. In doing so, we can achieve system-wide optimization on resource scheduling and collaboration for distributed edge environments.

随着网络边缘设备及数据量的爆炸式增长，以“计算应更靠近数据的源头”为核心理念的边缘计算模型应运而生。越来越多的数据处理作业涉及到跨边缘的数据汇聚和协同处理。为跨边缘数据处理作业分配不恰当的边缘资源会导致服务质量下降，因此需要系统化构建适应于异构边缘云的分布式资源调度框架。初步调研结果表明，跨边缘数据处理作业的资源需求在时间和空间维度上可能存在可利用的模式。这些发现促使我们去利用模式驱动方法指导资源预分配，构建拓扑感知的多维度边缘云综合评估体系，利用“用户-边缘云提供商-服务提供商”三方协作来优化资源调度，并进一步借助数据和任务复制实现关键作业保障，实现跨边缘云的统一资源调度和优化协同。

本项目研究面向跨边缘作业的分布式资源调度与协作框架。基于该项目，项目团队在智能化计算卸载、面向视频分析的边缘资源调度、服务实体部署与资源分配、基于边缘的神经网络推断加速等方面进行了深入的研究，取得了重要进展。发表或含录用论文34篇（含16/11篇CCF A/B类论文）；授权3项专利，申请3项专利；获得2021年度江苏省计算机学会青年科技奖、2020年度ACM南京分会新星奖、2019年度江苏省科学技术奖一等奖、国际会议ICPADS 2021最佳论文亚军奖、国际会议ICCCN 2020最佳论文奖等奖项。