Since intelligent algorithms such as deep learning (DL) have strong expressiveness, flexibility, and scalability for complex problems, some DL-based methods have been applied to Internet architecture-related scenarios, such as congestion control and malicious traffic detection. However, DL-based models have relatively high requirements for resources such as data and computing, and the existing Internet architecture requires further evolution to break through these limitations. In this article, we propose the collaboration-enabled intelligent Internet architecture, which can leverage intelligence to facilitate the evolution of Internet architecture in more complex scenarios. Specifically, we first discuss the inherent opportunities and challenges of enabling the Internet architecture to be intelligent through collaboration, which are brought by the imbalance of Internet supply and demand, distributed organizational structure, and the lack of built-in security. Immediately after, we present the newly proposed collaboration-enabled intelligent Internet architecture, which consists of heterogeneous hardware infrastructure and a collaboration-oriented software service platform. Through the complementarity of these two components (i.e., providing hierarchical computing and full exploitation of hierarchical capabilities), it promotes the collaboration of intelligent algorithms built into the Internet architecture. Moreover, some flexible algorithmic modules for the proprietary requirements of the Internet architecture are built into the software service platform. Finally, we take multi-classification malicious traffic detection as a case study, and demonstrate the advantages of enabling Internet architecture to be intelligent through collaboration.
Read full abstract