I-Corps: SDNatics: Big Data Analytics of Software Defined Networks to Understand, Predict and Protect Critical Computer Networks

传统的计算机网络管理在粒度、响应性和可预测性方面具有严重的限制，这可能会导致关键领域的网络故障，例如公共安全和医疗保健，造成数百万美元的损失。当今的网络目前是根据各种企业策略和最先进的工具(如软件定义网络(SDN)控制器)进行管理的。然而，网络中的故障和问题仍然会出现，因为总是存在预定义的规则和算法不能很好地响应的情况，导致流量泛滥、拥塞或易受攻击。深入了解网络行为模式并将知识应用于网络管理和规划是提供新的解决方案和最佳实践的基础。该团队开发了一套协议和分析方法，通过使用软件定义网络(SDN)和机器学习技术来了解网络动态。有了这些信息，该团队就能够报告异常网络行为并警告潜在故障。这允许以算法、自动和智能的方式管理更多的网络。具体地说，该团队的技术能够使用SDN技术跟踪更高粒度(即分辨率)的网络流量。然而，尽管当前的SDN方法在定义和执行围绕网络流的规则方面比传统网络更好，但它们没有能力分析深入的使用行为并动态地应用这些信息来改进网络管理和规划。通过使用机器学习技术(即深度学习)，该小组将开发SDNatics软件平台，以便更先进地使用SDN。这样的改进应该会提高网络系统的安全性、可预测性、速度和数据量。

Conventional computer network management has serious limitations on granularity, responsiveness and predictability, which can lead to network failures costing millions of dollars in critical domains, e.g., public safety and health care. Today's networks are currently managed under various enterprise policies and with state-of-the-art tools such as software defined networking (SDN) controllers. However, the faults and problems in the networks still arise because there are always cases to which predefined rules and algorithms do not respond well, causing traffic flooding, congestions, or vulnerability to attacks. Understanding in-depth of the network behavior patterns and applying the knowledge in network management and planning are fundamental to providing new solutions and best practices.This team has developed a suite of protocols and analytics methods to understand network dynamics by using software-defined networking (SDN) and machine learning technologies. With this information the team is able to report abnormal network behavior and alert potential faults. This allows for much more of the network to be managed algorithmically, automatically, and intelligently. Specifically, this team's technology is able to track network flows to a much higher granularity (i.e. resolution) using SDN technology. However, while current SDN approaches are better than traditional networks in defining and enforcing rules around network flows, they do not have a capability to analyze in-depth usage behavior and dynamically apply such information to improve network management and planning. By using machine learning techniques (i.e. Deep Learning), the team will develop the SDNatics software platform that will allow for more advance uses of SDN. Such improvements should improve security, predictability, speed, and data volume in network systems.