软件定义网络中应用与网络合作的资源分配模型及机制研究

Today’s Internet does not feature a direct information exchange between applications and the network. Due to the varying requirements of the heterogeneous applications running top of the network, this may result in a poor user-perceived quality of applications. Software-Defined Networking (SDN) provides a promising solution to this situation. By opening the northbound interface of SDN to applications, there are chances for sharing dynamic status information between applications and the network, which could be beneficial for both network resource management and application control. However, this could be a double-edged sword. For example, Each network-aware application tries to maximize their quality based on the available resources, but may also behave egoistically trying to supersede competing traffic. In this circumstance, giving authority of network resource allocation to applications could be a disaster. To address these issues, we argue that cooperative resource allocation between network and applications could be the way out. Based on this argument, we present novel control models for network resource allocation. In these models, we take all the decision-making entities, such as SDN controller and application controllers, into account. Using game theory and optimization theory, we can capture the interaction among these entities, predict the decisions been made, and analyze the optimality, stability, and the complexity of the decision-making process. Guided by the insights gained in these models, we look more deeply into two specific use cases, the hybrid SDN and the Data-Center networks, which can be used as typical examples of integrated control of applications and the network. Besides all the theoretic effort mentioned above, we plan to implement an appropriate experiment platform to evaluate the effects and performance of the control models.

现今的互联网中，网络与应用之间没有合适的接口来实现状态信息的相互交流。随着应用种类的增加，以及应用对网络资源的需求越来越复杂，这种缺乏交流的资源分配模式越来越不能满足要求。软件定义网络的出现有望改变这一现状。软件定义网络的北向接口可以成为应用和网络之间交流状态信息和控制命令的合适平台。但是，开放的北向接口也会给资源分配问题带来新的挑战。为应对这些挑战，本项目将“网络与应用合作的资源分配”作为核心理念，提出新的资源分配控制模型，以博弈论和优化理论为建模工具，探索各个决策实体之间的互动关系。并在此基础上着重解决资源分配决策过程中的稳定性、最佳性、复杂性等关键科学问题。本项目将以混合式软件交换网络和数据中心网络为研究案例，发展具体的资源分配方法，验证所提出模型的有效性。此外，本项目还将搭建适当规模的实验网络，为相关机制和机理提供研究和验证的平台。

本项目主要针对网络资源管控技术展开研究. 我们的核心理念是: 网络与应用必须合作, 才能保证资源配置的高效和灵活, 才能适应未来网络动态性大、应用需求复杂的要求.为此, 我们重点研究了三个方面的内容: 典型案例、联合优化方法以及网络测量关键技术. 在典型案例方面, 我们重点考察了数据中心的Cotask调度问题, 互联网中的DASH业务资源分配问题, 以及移动边缘计算中的任务卸载问题等. 这些案例中都需要网络层和应用层的相关信息才能做出最佳决策. 我们在这些问题上提出的联合优化解决方案在性能上有了显著提升. 验证了合作管控的合理性和有效性. 合作管控技术的困难主要有两个方面: 约束数目和求解空间的增大; 以及动态变化的条件. 为了应对这些挑战, 我们重点发展了三类联合优化求解方法: 有竞争比保证的在线算法、基于Lyapunov优化的在线算法, 以及基于机器学习的求解算法. 在VNF放置问题、MEC的任务卸载决策, 以及MEC资源分配问题这三个具体案例中, 我们分别应用上述求解方法, 性能得到了明显提升. 此外, 为实现合作管控, 对网络状态的精确感知和测量无疑是必须的. 但在网络动态变化, 且用于测量的资源有限的条件下, 如何实现准确、全面的测量是个有挑战性的任务. 本项目将这类问题建模为组合优化/凸优化问题, 并发展了启发式求解算法. 在满足资源约束的前提下, 提升了测量准确度..本项目成果有助于推动基于网络与应用合作的资源管控技术的应用和发展, 为更为高效、灵活的网络资源分配技术奠定了基础.