Thrift: optimizing performance in complex mobile internet applications with network coding

复杂互联网应用的用户，如增强现实导航，已经变得越来越移动，使用包括智能手机、平板电脑和超便携笔记本电脑在内的各种移动设备。这些应用程序需要来自公共或企业云计算平台的服务，以便共享应用程序数据、跨会话保持应用程序状态以及与其他用户协作。随着用户对这些应用提出更复杂的功能要求，以及存储和通信的负担转移到云上，需要云计算和互联网通信协议的集成设计，以满足用户对性能的期望。这项研究的目标是设计新的算法和协议，这些算法和协议针对移动互联网应用程序的性能进行了优化，并结合了新的移动互联网应用程序的延展性和云计算平台中数据中心丰富的存储和计算能力。 我们提出了Thrift，这是一个以无缝方式支持这种组合的新体系结构，并且网络编码作为我们设计的核心的一流原语被应用。我们采取整体观点，将数据中心、移动设备和之间的数据交付路径结合在一起，并针对影响性能的瓶颈和中断。Thrift架构监控涉及大量用户的移动应用程序中的重要性能指标，分析实时测量数据，使用高级统计和推断工具预测未来的资源需求，然后在云中和数据交付路径上做出分配资源和应用网络编码的决策。作为一个涉及真实世界用户的试验床，将在iOS移动平台上开发一个协作多媒体应用程序。随着加拿大人越来越依赖移动设备和在这些设备上运行的应用程序，我们拟议的工作在优化各种此类移动应用程序的用户体验方面取得了重大进展。

Users of complex Internet applications, such as navigation with augmented reality, have become increasingly mobile, using a wide range of mobile devices, including smartphones, tablets, and ultraportable notebook computers. These applications require services from public or enterprise cloud computing platforms, in order to share application data, keep application states persistent across sessions, and collaborate with other users. As users demand more complex functionality from these applications, and as the burden of storage and communication is shifted to the cloud, an integrated design of both cloud computing and Internet communication protocols is required in order to meet user expectations on performance. The objective of this research is to design new algorithms and protocols that are optimized for performance in mobile Internet applications, and that combine the malleability of new mobile Internet applications and the availability of abundant storage and computation power in data centres in cloud computing platforms. We propose Thrift, a new architecture that supports such a combination in a seamless fashion, and network coding is applied as a first-class primitive at the core of our design. We take a holistic view that combines data centres, mobile devices, and data delivery paths in between, and target bottlenecks and disruptions that throttle the performance. The Thrift architecture monitors important performance metrics in a mobile application involving a large number of users, analyzes live measurement data, predicts future resource demand using advanced statistical and inference tools, and then makes decisions that allocate resources and apply network coding, both in the cloud and along paths of data delivery. As a testbed involving real-world users, a collaborative multimedia application will be developed on the iOS mobile platform. As Canadians increasingly depend on mobile devices and applications running on these devices, our proposed work makes significant advances towards optimizing the user experience in a wide variety of such mobile applications.