NeTS: Medium: Collaborative Research: Tango: Performance and Fault Management in Cellular Networks through Device-Network Cooperation

蜂窝网络已经成为我们赖以生存的基础设施的一部分，我们没有停下来思考这种网络背后的巨大复杂性。这些网络服务于种类繁多的移动设备，并在种类繁多的无线电环境中运行。这样的网络故障并不罕见，我们把它们的后果当作生活中的事实，抱怨掉线和数据中断。同样，我们也会遇到性能下降的问题，要么是通话的语音质量下降，要么是连接速度下降。随着蜂窝网络通信量的快速增长和蜂窝频谱的紧张，这些情况预计将在近期和中期变得更加严重。此外，在今天的网络中，没有设施可以通过编排地理上分散的移动设备的整个网络来为设备提供服务，例如提供迎面而来的龙卷风路径的“视频”。在这个项目中，研究人员提出了一种观点，即需要在蜂窝网络中建立主动管理此类事件，而不是被动地管理故障或性能下降。这样做的主要困难在于，对于终端设备来说，网络本质上被视为一个黑匣子。关于网络的条件，例如蜂窝基础设施的某些元素的拥塞或过载，对移动终端设备上的应用程序不可见。同样，设备内的大部分应用程序生态系统对网络是不可见的。因此，网络不知道某些应用程序的延迟需求是什么，或者在不久的将来它将在网络上放置什么需求。提出的解决方案将彻底改变这种黑盒方法，并通过网络和设备之间的合作实现主动故障和性能管理，以及一种全新的应用程序类别，这是一种探戈。项目目标将通过两个协同的项目推进来实现。在第一个名为“故障管理和编排服务”的项目中，该团队将为实时蜂窝数据开发计算和统计模型，并为此开发分析工具、预测硬故障和软故障的机器学习算法，以及避免预测故障的缓解措施。它还将开发蜂窝网络和最终用户之间的协调服务，以在“参与式感知”的广泛保护伞下实现全系统目标。第二个重点是“性能驱动设计”，该团队将开发一个API，允许蜂窝设备和蜂窝网络合作，以实现共同期望的系统级属性，例如更高的数据带宽。它将提供先进的流量管理方案，通过回填可容忍延迟的流量和从移动设备上卸载计算，利用团队成员最近在卷云项目中的努力。该项目将开发一个微型LTE/UMTS测试平台，用于开放实验、开放数据访问和对上述协议的评估。

Cellular networks have become a part of the infrastructure that we rely on without pausing to think of the enormous complexity that underlies such networks. These networks serve an amazing diversity of mobile devices and operate in an amazing diversity of radio environments. Failures in such networks are not uncommon and we take their consequences in our stride as a fact of life, complaining about call drops and data disconnections. Similarly, we also encounter performance degradations, either in voice quality of calls or connection speeds. With the fast increasing amount of traffic on cellular networks and the cellular spectrum crunch, these conditions are expected to become more severe in the near to mid-term future. Further, in today's networks, there is no facility to provide services to a device by orchestrating an entire network of geographically dispersed mobile devices, such as, for providing a "video" of the path of an oncoming tornado.In this project, the researchers put forward the view that rather than reactive management of failures or performance degradations, there is a need to build into cellular networks, proactive management of such events. The primary difficulty for this is that, to the end devices, the network is considered essentially as a black box. Conditions about the network, such as, congestion or overload of some elements of the cellular infrastructure, are not made visible to the applications on the mobile end devices. Likewise, much of the application eco-system within a device is not made visible to the network. Thus, it is not known to the network what is the latency requirement of some application or what demand it is going to place in the near future on the network. The proposed solution approach will turn this black box approach on its head and enable proactive fault and performance management and an entirely new class of applications through cooperation between the network and the devices, a Tango of sorts.The project goals will be accomplished through two synergistic project thrusts. In the first, called "Fault Management and Orchestrated Services", the team will develop computational and statistical models for the real-time cellular data and analytics tools for the same, machine learning algorithms for predicting hard and soft failures, and mitigation actions for avoiding predicted failures. It will also develop orchestrated services between the cellular network and the end users to achieve system-wide goals under the broad umbrella of "participatory sensing". In the second thrust, called "Performance-driven Design", the team will develop an API that will allow the cellular devices and the cellular network to cooperate for achieving common desired system-level properties, such as, higher data bandwidth. It will provide advanced traffic management schemes through back-filling of delay-tolerant traffic and computation offloading from the mobile devices, leveraging a team member's recent efforts in the Cirrus Clouds Project. The project will result in the development of a miniature LTE/UMTS testbed for open experimentation, open data access, and evaluation of the protocols mentioned above.