Fog Machine: A Fault-Tolerant System and Programming Framework for Cloud of Moving Things

云计算和移动的计算是过去几年计算趋势的两个主要驱动力。使用云计算为移动的应用提供动力的挑战之一是服务器资源和客户端之间的显著网络距离以及在交付有保证的服务质量方面随之产生的问题。为了解决这些问题，人们提出了“边缘”计算范例，例如使云更接近客户端的雾。随着物联网（如联网汽车和智能家居）的蓬勃发展，边缘计算将在未来的任何计算基础设施中发挥更大的作用。由这种云-雾-设备系统形成的三层分布式计算架构引入了许多新的挑战。其中最主要的是系统的可编程性。* 本研究的第一个目标是为三层云-雾-设备的移动物联网（CoMT）系统开发一个模型，该模型提供（a）基于设备位置的自适应设备-雾关联，（B）跨雾的移动的设备的有效切换，（c）满足可靠性目标的系统重新配置。第二个目标是开发一个简单而灵活的编程模型，允许程序员控制CoMT系统中故障管理，一致性和性能的各个方面。* 雾用于提供对处理能力的低延迟和本地化访问。出于容错目的，我们需要将多个雾与设备关联，以便在雾或网络故障时可以快速故障转移。与设备相关的雾需要仔细选择，以最大限度地减少延迟，并且应该来自适当的位置。对于移动性，情况可能会变得非常复杂;设备需要更改关联以保持本地性和低延迟访问。在像车辆网络这样的快速移动场景中，从一个雾到另一个雾的切换时间需要非常小。在CoMT中，设备可以调用雾或云对它们捕获的数据执行各种类型的处理。因此，由设备、雾和云组合而成的计算系统需要可靠，并以一致的方式维护状态。* 增加容错的一个简单方法是在系统中插入冗余。在三层CoMT的情况下，我们可以在设备和雾之间以及雾和云之间建立冗余关联。然而，建立太多这样的联系是有代价的。特别是，完成基本操作的时间（例如，读和写）随着FOG数量的增加而增加。因此，需要在增加冗余的好处和增加维护一致机器的成本之间进行仔细的权衡分析。

Cloud computing and mobile computing have been two dominant drivers of computing trends in the last few years. One of the challenges of using cloud computing to power mobile applications is the significant network distance between the server resources and the clients and the consequent problems in delivering guaranteed qualities of services. To address these problems “edge” computing paradigms such as fogs that bring clouds closer to the clients have been proposed. With the impending boom of Internet of Things such as connected cars and smart homes, edge computing is poised to play a larger role in any computing infrastructure of the future. The three-layered distributed computing architecture formed by such a cloud-fog-device system introduces many new challenges. Chief among them is the programmability of the system. ***The first objective of this research is to develop a model for Cloud of Moving Things (CoMT) system for the three-layered cloud-fog-device that provides (a) adaptive device-to-fog associations based on the locality of the devices, (b) efficient handover of mobile devices across the fogs, (c) system reconfiguration to meet the reliability goals. The second objective is to develop a simple yet flexible programming model that allows the programmer to control various aspects of failure management, consistency, and performance in a CoMT system. ***Fogs are used to provide low latency and localized access to processing capacities. For fault tolerance purposes, we need to associate multiple fogs with a device so that quick failovers are possible with fog or network failures. The fogs that are associated with a device needs to be selected carefully to minimize latencies and should be from the appropriate locality. With mobility, the situation can get quite complicated; the devices need to change associations to maintain the locality and low latency access. In a rapidly moving scenario like vehicular networks the handover times from one fog to another need to be very small. In CoMT, the devices can call upon the fog or cloud to perform various types of processing on data they have captured. Therefore, the computing system formed by the combination of devices, fogs, and clouds needs to be reliable and maintain state in a consistent manner. ***A simple way to add fault tolerance is to insert redundancy into the system. In the case of the three-layered CoMT, we can make redundant associations between the devices and fogs and between fogs and clouds. However, making too many such associations come with a cost. In particular, the time to complete the basic operations (e.g., reads and writes) on fogs increase with increasing number of fogs. Therefore, a careful trade-off analysis need to be performed between the benefits of increasing redundancy and increasing costs of maintaining a consistent machine.