Future Internet is expected to meet explosive traffic growth and extremely complex architecture, which tend to make the traditional NTC strategies inefficient and even ineffective. Inspired by the latest breakthroughs of AI and its power to address large-scale and complex difficulties, the network community has begun to consider shifting the NTC paradigm from legacy rule-based to novel AI-based. As an applied inter-discipline, design and implementation are important. Although there have been some preliminary explorations along this frontier, they are either limited by only envisioning the prospects, or too scattered to provide high-level insight into a general methodology. To this end, we start with the domain knowledge relationships of AI and NTC, summarizing a baseline workflow toward deep reinforcement learning, which will be the dominant method for the AI-NTC paradigm. On top of that, we argue that AI-NTC training and running must be carried out in online environments in closed-loop fashion for the purpose of putting ti into practice. A series of challenges and opportunities are discussed from a realistic viewpoint, and a set of new architecture and mechanism to enable the online and closed-loop AI-NTC paradigm are proposed. Hopefully, this work can help the AI community to better understand NTC and the NTC community to better live with AI.

中文翻译：

---

与人工智能共存：未来网络流量控制的范式转变


未来的互联网预计将面临爆炸性的流量增长和极其复杂的架构，这往往会使传统的NTC策略变得低效甚至无效。受人工智能最新突破及其解决大规模复杂困难能力的启发，网络社区已开始考虑将 NTC 范式从传统的基于规则的模式转变为基于新型人工智能的模式。作为一门应用交叉学科，设计和实施非常重要。尽管在这一领域已经进行了一些初步探索，但它们要么仅限于展望前景，要么过于分散，无法提供对通用方法的高层次洞察。为此，我们从AI和NTC的领域知识关系入手，总结了深度强化学习的基线工作流程，这将成为AI-NTC范式的主导方法。除此之外，我们认为AI-NTC的训练和运行必须在在线环境中以闭环方式进行，才能将TI付诸实践。从现实的角度讨论了一系列挑战和机遇，并提出了一套新的架构和机制来实现在线和闭环的AI-NTC范式。希望这项工作能够帮助 AI 社区更好地了解 NTC，并帮助 NTC 社区更好地与 AI 共存。