SPINE: Stochastic Performance Bounds for Information-Centric Communication Networks

内容高速缓存和分组转发功能的集成是通信网络中的最新发展步骤，这是联网的存在理由从连接主机转变为提供内容的结果。尽管最近兴起了具有缓存能力的网络，例如以信息为中心的网络(ICN)的概念，作为未来互联网体系结构的候选，但缓存和转发元素交互的基础仍然没有被很好地理解。大多数方法基于当前建立的内容交付体系结构，该体系结构不捕获网络层上的缓存和排队交互。对于ICN，传统的业务流和端到端连接概念并不成立，这些方法引入了无法用这些方法解释的性能建模错误。在这个项目中，我们的目标是推导出一种基于排队论的方法，用于利用随机通信系统的以信息为中心的网络的性能分析和自适应控制。这样的系统包括支持缓存的路由器和时变可用带宽资源和突发流量下的调度器。我们期望来自排队论和随机网络演算的公式，例如Fork-Join模型和统计服务曲线，有可能产生新的见解，例如，关于ICN调度算法的最佳性，这可以导致ICN中的应用定制通信方案。我们预计，自适应控制方法，如马尔可夫决策过程，与统计推理算法相结合，可以深入了解自适应ICN通信的基本性能限制。这一建议的预期结果将提供对ICN设计空间的基本理解，并使ICN协议和应用能够进行超出当今理论最先进水平的严格优化，例如使得能够针对具有马尔可夫可加交叉业务的ICN业务实现最优ICN分组调度。

The integration of content caching and packet forwarding functionalities is a recent evolutionary step in communication networks that is a consequence of a shift in the raison d'être of networking from connecting hosts to delivering content. Despite the recent rise of caching-capable networks, such as the concept of Information-Centric Networking (ICN), as candidates for future Internet architectures, the foundations of the interactions of caching and forwarding elements are still not well understood. Most methods are based on currently established content-delivery architectures that do not capture caching and queueing interactions on the network layer. For ICN, where, in addition, traditional notions of traffic flows and end-to-end connections do not hold, such methods introduce performance modeling errors that cannot be explained by these approaches.In this project, we aim at deriving a queuing theory based methodology for the performance analysis and adaptive control of information-centric networks that utilize stochastic communication systems. Such systems include caching-capable routers and schedulers under time-varying available bandwidth resources and bursty traffic. We expect that formulations from queueing theory and stochastic network calculus, such as Fork-Join models and statistical service curves, have the potential to yield new insights, e.g., regarding the optimality of ICN scheduling algorithms, which can lead to application-tailored communication schemes in ICN. We anticipate that adaptive control methods, such as Markov decision processes, paired with statistical inference algorithms allow gaining insights on fundamental performance limits for adaptive ICN communications. The expected results of this proposal will provide a fundamental understanding of the design space of ICN and enable rigorous optimizations of ICN protocols and applications that are beyond the state-of-the-art of today's theories, for example enabling optimal ICN packet scheduling for ICN traffic with Markov-additive cross traffic.