绿色边缘计算网络资源优化配置方法研究

Networkization is the key to the success of cloud computing and big data. With the wide application of Internet of Things, the amount of traffic as well as the number of flows is extremely massive with variety, and the service requirements of flows become individualized. In this sense, the flow’s quality of service significantly affects its user’s quality of experience. However, the quality of service in industrial data center networks is now provisioned at the cost of network resource over-allocation, which wastes lots of network and energy resources, resulting in high operating cost. The edge computing network, which joints edge computing and cloud computing together, has recently appeared as a new network paradigm for network efficiency. However, the detail technologies of how to efficiently utilize the limited network resource to provision quality of server deserve further study. Firstly, it is a challenge to handle the problem of allocating network resource to massive flows with distinct quality of service requirements by using limited rules. Secondly, the service requirement of a flow varies dynamically. Furthermore, the resource requirements change fast. In addition, the heterogeneous computing capability as well as energy consumption between edge nodes and clouds complicates the problem. This project is conducted to address resource allocation problem in edge computing networks by proposing a dynamic and cost effective network resource optimal allocation scheme, including energy-efficient flow splitting with dynamical edge computing offloading, self-adaptive and fast packet scheduling and flow access control. Our scheme explicitly defines and determines the service requirement of every flow, predicts the trends of traffic change, and optimizes resource allocation with multiple technologies. It is expected that the research of this project will satisfy the quality of service requirements of massive flows while reducing energy consumption and operating cost, and lead to green edge computing networks.

网络化是云计算、大数据等新兴技术成功的关键。随着物联网应用的普及，网络流量愈加海量化、服务要求更加个性化。每个流的服务都直接影响着用户／物的满意度。目前的做法主要是通过网络资源过度配置的方式来满足不同的流的服务要求，往往造成网络资源和能耗的浪费。边缘计算和云计算结合的边缘计算网络范式将是提高网络资源效率的有效方法，但边缘计算网络的绿色化尚待深入研究。首先，用非常有限的规则对海量的不同服务要求的流进行资源配置已是一个非常具有挑战性的问题；其次，每个流的要求千变万化；再者，资源需求瞬时变化；还有，节点计算能力、能耗的异构性也限制了问题的解。本项目拟从节能边缘分流、快速分组调度和流接入控制这三个方面展开研究，提出一种动态的、节能的网络资源优化配置方法，明确每个流的要求、预测流量变化和跨层多技术协作优化资源配置，既满足不同的流的服务要求又降低能耗和运营成本，为实现绿色边缘计算提供理论依据。

随着物联网应用的普及，网络流量愈加海量化、服务要求更加个性化。每个流的服务都直接影响着用户/物的满意度。边缘计算和云计算结合的边缘计算网络范式将是提高网络资源效率的有效方法。但是，如何快速高效地利用异构的端边云计算资源、有限的网络带宽资源来满足数据流千变万化的需求和瞬时资源需求，同时又降低能耗和运营成本，是边缘计算网络绿色化面临的一个重要问题。针对这一问题，本项目进行了以下三个方面的研究：（1）针对网络边缘计算资源、网络带宽资源的受限性和实时业务流量动态变化特性，进行了网络主导的节能分流策略研究，提出了联合缓存管理、队列调度的计算资源、网络资源协同优化的计算迁移策略，从流量负载的角度来提高网络的服务能力；（2）为了对流量瞬时突变作出毫秒量级的网络响应，提出了基于延迟的分组包快速调度和缓存接入控制方法；（3）针对流量突发引起网络丢包的问题，研究了分流与分组调度协助的流接入控制，提出了基于动态权重滑动窗口的流接入控制方法。上述研究既满足进入网络的分组包的区分延迟服务要求又降低了能耗和运营成本。课题按项目预定的计划进行，达到了项目预期的目标，并完成了项目预期的研究成果。目前，形成的研究成果包括：（1）在国内外高水平学术期刊和国际会议上发表和录用相关论文17篇，其中，SCI源刊9篇，EI检索7篇；（2）申请国内发明专利4件，已授权2件；（3）培养研究生5名。