Abstract We study a future Internet networking paradigm where instead of congestion control an open loop traffic control is applied. We aim to give theoretical foundations for data transfer controlled only by the access points of the network. The key characteristics of networks without congestion control are stability and efficiency addressed in this paper. We consider the queue length processes of data-flows on directed graphs. The stability is characterized by the ergodicity of these processes and the efficiency of the network is measured by the Price of Anarchy. Under restrictions on the input traffic rates we derive an achievable efficiency limit in a stable network for very general conditions, namely, for any network topology and for any buffer management policy. Moreover, we show that even for cyclic networks, which usually cause severe instability in networks, an upper bound for the loss of efficiency can be given independently of the size of the network under a fair AQM buffer management policy. Furthermore, for monotonic networks we present a reasonable choice for setting access capacities. Our results demonstrate that with a proper setting of access capacities of incoming flows the congestion collapse of the Internet can be avoided even without congestion control.

中文翻译：

---

没有拥塞控制的网络

摘要 我们研究了未来的互联网网络范式，其中应用了开环流量控制而不是拥塞控制。我们旨在为仅由网络接入点控制的数据传输提供理论基础。没有拥塞控制的网络的关键特征是本文中讨论的稳定性和效率。我们考虑有向图上数据流的队列长度过程。稳定性的特点是这些过程的遍历性，网络的效率由无政府状态的价格来衡量。在输入流量速率的限制下，我们得出了一个稳定网络中非常普遍的条件下可实现的效率限制，即对于任何网络拓扑和任何缓冲区管理策略。此外，我们表明，即使对于循环网络，这通常会导致网络严重不稳定，在公平的 AQM 缓冲区管理策略下，可以独立于网络的大小给出效率损失的上限。此外，对于单调网络，我们提出了设置访问容量的合理选择。我们的结果表明，通过适当设置传入流的访问容量，即使没有拥塞控制，也可以避免互联网的拥塞崩溃。