面向复杂多样物联网业务的边缘计算资源优化配置与部署研究

With the continuous emergence of complex and diverse services in Internet of Things (IoT), how to provide high quality edge computing services for differentiated applications has important theoretical and practical significance. Aiming at IoT applications with the characteristics of large data quantity, strong delay sensitivity, high computing intensity and heterogeneity of multi-layer device-edge-cloud resources, this project studies optimal resource allocation and deployment of edge computing for varied complex services in IoT in order to improve system performance, quality of services and user experience. The quality evaluation function of component clustering is defined. The graph mapping between application graph and resource graph is built. The context-aware fine-granularity self-adaptive task offloading method in edge computing is proposed to reduce cost and energy consumption. Subsequently, the Voronoi diagram is adopted to divide edge service areas, and cost function of edge-cloud resource allocation is constructed. The smooth dynamic allocation strategy for multi-layer heterogeneous resources is suggested to minimize time delay and resource allocation cost. Moreover, the joint utility function based on cache value, content reliability and network cost is constructed. The distributed cooperative edge cache scheduling algorithm is designed. Furthermore, the sparse deployment optimization strategy of edge servers based on user distribution and node costs is studied to provide efficient edge services. Finally, the edge computing experimental environment for Internet of Vehicles and intelligent manufacturing is built to test the proposed methods, and the demonstration applications of campus edge computing is also provided.

随着复杂多样的物联网业务不断涌现，如何为差异化的应用提供优质的边缘计算服务具有重要的理论和实际意义。本项目针对物联网应用数据量大、时延敏感性强、计算密集度高以及“终端-边缘-云”多层资源的异构性等特点，研究面向复杂多样物联网业务的边缘计算资源优化配置与部署，以提升系统性能、边缘服务质量及用户体验。定义组件聚类质量评价函数，建立应用图和资源图的映射，提出边缘计算中上下文感知的细粒度自适应任务卸载方法，以降低成本及能耗。采用维诺图划分边缘服务区域，构建“边缘-云”的资源配置成本函数，建议多层异构资源平滑动态分配策略，最小化时延及资源配置成本。建立基于缓存价值、内容可靠性和网络代价的联合效用函数，设计分布式协同边缘缓存调度算法。研究基于用户分布和节点代价的边缘服务器稀疏部署优化策略，以提供高效的边缘服务。构建面向车联网和智能制造的边缘计算实验环境，测试所提出的方法，并提供校园边缘计算示范应用。

本项目针对物联网应用时延敏感性强、计算密集度高以及“终端-边缘-云”三层资源的异构性等特点，多角度研究面向复杂多样物联网业务的边缘计算资源优化配置与部署。如何选择高效的卸载策略，扩展用户终端的处理能力，满足大规模实时物联网应用需求是亟待解决的问题。针对频繁更新的边缘计算上下文环境，本项目综合考虑用户终端和边缘计算环境中卸载的综合代价，提出了边缘计算环境中多组件应用计算卸载策略。实验结果表明所提方法能有效解决终端设备计算性能以及电量有限的问题，并协同利用邻居和远程云资源，处理多组件的物联网应用作业，减少作业完成时间，降低用户设备终端能耗，从而提升用户体验。在终端设备能耗和边缘服务器性能等多因素约束下，如何合理利用云端、边缘服务器以及终端设备组成的分布式多层异构资源，在满足应用实时性要求的同时最小化资源配置成本是边缘计算需要关注的问题。本项目建议了能耗成本感知的“边缘-云”多层异构资源动态分配方法。实验结果表明，在能耗和SLA违规方面，所提出的方法明显优于所比较的方案。边缘计算环境中单个边缘服务器存储能力有限，层次式缓存机制导致延迟较长并浪费存储空间，只考虑单个因素的缓存内容调度策略会导致缓存价值降低和带宽消耗代价增加。针对以上问题，本项目提出边缘计算中分布式协同边缘缓存放置算法。本项目所提出的协同缓存系统数据放置策略能高效的搜索流行度高的数据最佳的放置位置，提高缓存命中率，降低回传网络传输流量。如何为边缘服务器选址并确定该位置节点数量以实现用户低延迟需求和节点利用率均衡是亟待解决的问题，本项目建议了成本感知的边缘服务器优化部署方法。实验结果表明，该策略可以更加准确地规划出边缘的位置及每个边缘服务器的数量，可以保障用户低延迟应用的需求，并能提高资源的利用率从而使整体边缘服务器部署成本较低。