Applying Learning Theory to Networking Problems

这个项目将专注于将学习理论研究人员开发的成果应用于现实生活中的问题。这种以应用为基础的理论工作很重要，以便学习模型和所研究的问题尽可能最好地捕捉到现实生活中问题的需求。努力将理论技术应用于实际问题，可以更好地理解现实世界的问题，从而有助于指导未来的理论工作，促进成果从理论转化为实践。许多联网协议(以及其他系统控制算法)利用一个或多个可调参数，例如阈值和窗口大小。通常，用于设置或调整参数值的方法是特别的，并且基于关于操作环境的某些假设，例如，假设通信网络中的业务模式的特定分布。本项目将开发基于形式化学习方法的框架，用于系统控制算法中的自动参数整定，以研究是否可以获得更好的性能。本项目将研究的一种这样的网络应用是动态调整TCP协议中确认的延迟。延迟确认有两个主要优点。首先，它允许对多个数据包进行一次确认。其次，如果数据包以相反的方向发送，那么我们可以利用确认，太多可能会增加延迟。目前使用的大多数TCP实现都采用了某种确认延迟机制。本项目研究了几种动态调整确认延迟的方案，包括基于加权多数(WM)算法的方案和基于指数加权移动平均(EWMA)算法的方案。这些方案将在效率和有效性的基础上进行对比，并根据它们对“不稳定”的TCP连接的响应程度来证明有关它们的理论结果。也就是说，如果连接偶尔从稀疏数据传输(其目标可能是最小化延迟，如在交互式应用的情况下)切换到密集数据传输(其目标可能是最小化确认的数量，并且延迟并不像文件传输的情况那样重要)。仿真结果将作为理论结果的补充。将学习理论结果应用于这个问题和其他网络问题，将为定义新的理论学习模型提供指导，这些模型可以更好地模拟现实生活场景。该项目还将认真开发和研究这种新的学习模式。

This project will focus on applying results developed by learning theory researchers to real-life problems. Such application-based theoretical work is important so that the learning models and problems studied capture as best possible the needs of real-life problems. Working to apply theoretical techniques to real problems creates a better understanding of the real-world problems and thus can help direct the future theoretical work, facilitating the transfer of results from theory to practice. Many networking protocols (as well as other system control algorithms) utilize one or more tunable parameters, e.g. thresholds and window sizes. Frequently the methods used to set or adjust the values of the parameters are ad hoc and are based on certain assumptions about the operating environment, e.g. assuming a particular distribution on the traffic patterns in a communication network. This project will develop framework based on formal learning methods for automatic parameter tuning in system control algorithms to study if better performance can be obtained. One such networking application this project will study is that of dynamically adjusting delays of acknowledgements in the TCP protocol. Delaying acknowledgements has two main advantages. First, it allows a single acknowledgement for more than one packet. Second, if a data packet is being sent in the opposite direction, then we can piggy-back the acknowledgement too much can increase the latency. Most TCP implementations used today employ some sort of acknowledgement delay mechanism. This project investigates several schemes to dynamically adjust acknowledgement delay, including ones based on the Weighted Majority (WM) algorithms and ones based on the Exponentially Weighted Moving Average (EWMA) algorithm. These schemes will be contrasted based on efficiency and efficacy , as well as proving theoretical results about them in terms of how well they respond to "volatile" TCP connections. That is, if the connecti on occasionally switches from sparse data transmission (where the goal might be to minimize latency, as in the case of an interactive application) to dense data transmission (where the goal might be to minimize the number of acknowledgements and latency is not as important, as in the case of a file transfer) and back again. Simulation results will be used to supplement the theoretical results. Applying learning theory results to this problem and other networking problems will then provide guidance in defining new theoretical learning models that better model real-life scenarios. This project will also carefully develop and study such new learning models.