Orchestration and Management of Softwarized Networks

网络管理是监视和控制网络的过程，以便它继续有效地提供服务。传统的网络管理严重依赖于人工操作员，他们成本高，容易出错，并且适应变化缓慢。它无法科普由于网络设备数量和多样性的增加而带来的复杂性，以及新兴应用的严格要求。 几十年来，自动化管理一直是网络管理研究的圣杯。它旨在实现自治网络，即，网络能够自主地监视它们的状态、分析问题、做出决策和执行校正动作。尽管过去曾多次尝试实现自主网络，但它们的实际部署在很大程度上仍未实现。有几个因素导致了这一点，包括存在许多利益相关者，他们的目标相互冲突，依赖专有硬件，无法大规模处理网络监控数据，以及缺乏全局可见性限制了网络范围的优化。 由于（i）网络软件化的进步（一种新兴的范式，其中软件控制网络流的处理，通过软件处理为这些流增加价值，并编排网络服务以满足应用需求）;（ii）机器学习的最新突破;以及（iii）大规模数据处理平台的可用性，现在星星已经对准实现自主网络的愿景。然而，为了在这些不同的技术领域之间建立协同作用，并实现软件化网络的自动编排和管理，必须解决一些根本性的挑战。 该研究计划的长期目标是解决实现自主网络愿景的关键管理挑战。具体而言，我和我的团队将追求三个主要目标： 利用网络软件化和大规模数据流开发：（i）以最小开销收集网络数据的高效算法;以及（ii）可定制的监控探头，用于根据不断变化的网络条件检测更广泛的网络事件。 利用大规模机器学习的最新创新技术对监控数据进行实时分析，以提取知识，推断网络随时间的行为，并自动做出管理决策。 利用优化技术和面向服务的架构开发新的算法和系统，实现高效、弹性和动态的网络服务编排。 实现自主网络的愿景将有助于减少网络运营商的资本和运营支出，消除网络中断，优化网络运营，并更快地部署创新服务。拟议的研究计划将培养推动即将到来的网络创新浪潮所需的受训人员，并确保加拿大继续发挥领导作用。

Network management is the process of monitoring and controlling a network so it continues providing its services efficiently. Traditional network management heavily relies on human operators, who are costly, error-prone, and slow to adapt to changes. It cannot cope with the complexity due to the increasing number and diversity of network devices, and the stringent requirements of emerging applications. Automated management has been the holy grail of network management research for decades. It aims at achieving autonomous networks, i.e., networks capable of autonomously monitoring their status, analyzing problems, making decisions, and executing corrective actions. Despite several attempts to achieve autonomous networks in the past, their practical deployments have largely remained unrealized. Several factors contribute to this, including the existence of many stakeholders with conflicting goals, reliance on proprietary hardware, the inability to process network monitoring data at scale, and the lack of global visibility restricting network-wide optimizations. The stars are now aligned to realize the vision of autonomous networks thanks to (i) advances in network softwarization (an emerging paradigm where software controls the treatment of network flows, adds value to these flows by software processing, and orchestrates network services to meet application needs); (ii) recent breakthroughs in machine learning; and (iii) the availability of large-scale data processing platforms. However, a number of fundamental challenges must be addressed in order to create the synergy between these different technology domains and achieve automated orchestration and management of softwarized networks. The long-term goal of this research program is to address key management challenges for realizing the vision of autonomous networks. Specifically, my team and I will pursue three primary objectives: Leverage network softwarization and large-scale data streaming to develop: (i) efficient algorithms for collecting network data at minimal overhead; and (ii) customizable monitoring probes for detecting a wider range of network events based on changing network conditions. Leverage recent innovations in large-scale machine learning for real-time analytics of monitoring data to extract knowledge, infer the behavior of the network over time, and automatically make management decisions. Leverage optimization techniques and service-oriented architectures to develop novel algorithms and systems for efficient, resilient, and dynamic orchestration of network services. Realizing the vision of autonomous networks will contribute to reducing network operators' capital and operational expenditures, eliminating network outages, optimizing network operations, and deploying innovative services faster. The proposed research program will produce the trainees required to drive the coming wave of network innovation and ensure Canada will continue in its leadership role.