CAREER: Adaptive Web Execution: Supporting Billions of Diverse Users by Adapting Execution to Available Resources

Web页面提供对许多关键服务（例如，卫生保健、教育、新闻）的访问，并且越来越多地由拥有不同网络和设备资源的用户访问。不幸的是，尽管页面加载的性能和功能在不同的资源设置中会有很大的不同，但目前的页面加载对其执行环境的适应程度很低。这将导致资源未充分利用或页面损坏（不完整）。该项目旨在开发一种新的web范例，称为自适应web执行（AWE），其中页面加载直接根据可用资源调整其执行或内容。关键目标是根据用户的资源可用性最大化网页可以提供给用户的性能和功能。该项目涉及三个协同方向，开发实现AWE范式的基础算法和实用系统。首先，它将制定策略，以一种平衡跨堆栈分析开销和可操作的适应性见解的方式，收集和向页面加载公开资源信息。其次，它将设计一套内容保存优化系统，1)通过高效的、以网络为中心的机器学习动态地适应现有的优化策略，2)明智地将新的、未使用的设备资源纳入页面加载。第三，它将创建方法来简化内容修改适应性的开发，例如web感知重放调试，它可以全面评估跨潜在执行环境的页面修改。这项研究将从根本上改变和改善多方参与的网络。发达地区的用户将体验到更低的延迟或增加的功能（增加网站收入），而发展中地区的用户将获得适当的访问关键应用程序的权限，这些应用程序目前无法使用其可用资源。AWE还将降低网站的开发成本，并通过泛化优化来弥合网络研究与实际部署之间的差距。这项研究将由发展中地区的试验台提供信息并在试验台进行评估，并通过与工业合作者的伙伴关系进行评估。该项目还将涉及1)一种新的教学“全栈”方法来教授网络系统设计，以及2)努力通过网络的高度可访问性来吸引代表性不足的本科生和K-12学生。作为该项目的一部分，研究成果和课程材料将在一个公共网站上发布：https://web.cs.ucla.edu/~ravi/awe/。此外，项目站点将包括在开发区域测试台上收集的页面结构和资源可用性的汇总摘要（以保护隐私的模拟跟踪的形式）。网站将定期维护，项目数据将在发布后至少保留5年，并根据公众兴趣延长。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Web pages provide access to many critical services (e.g., health care, education, news), and are increasingly accessed by users with diverse network and device resources. Unfortunately, despite the fact that performance and functionality of page loads can vary drastically across resource settings, page loads today minimally adapt to their execution environments. This results in either underutilized resources or broken (incomplete) pages. This project aims to develop a new web paradigm called Adaptive Web Execution (AWE), in which page loads directly adapt their execution or content according to the available resources. The key goal is to maximize the performance and functionality that a web page can offer a user based on that user’s resource availability.The project involves three synergistic directions that develop the foundational algorithms and practical systems for realizing the AWE paradigm. First, it will develop strategies to collect and expose resource information to page loads in a way that balances cross-stack profiling overheads with actionable adaptation insights. Second, it will design a suite of content-preserving optimization systems that 1) dynamically adapt existing optimization strategies via efficient, web-focused machine learning, and 2) judiciously incorporate new, unused device resources into page loads. Third, it will create methods to simplify the development of content-altering adaptations, such as web-aware replay debugging that comprehensively evaluates page modifications across potential execution environments.This research will fundamentally transform and improve the web for multiple players. Users in developed regions will experience lower delays or increased functionality (increasing website revenue), while users in developing regions will get proper access to critical applications that are currently unusable with their available resources. AWE will also reduce development costs for websites and bridge the gap between web research and practical deployments by generalizing optimizations. The research will be informed by and evaluated in testbeds in developing regions, as well as through partnerships with industrial collaborators. The project will also involve 1) a new pedagogical 'full stack' approach to teaching networked system design, and 2) efforts to attract underrepresented undergraduate and K-12 students through the highly accessible lens of the web.The research artifacts and course materials designed as part of this project will be released on a public website: https://web.cs.ucla.edu/~ravi/awe/. In addition, the project site will include aggregate summaries of page structures and resource availability (in the form of privacy-preserving emulation traces) collected on developing region testbeds. The site will be regularly maintained, and project data will be kept for at least 5 years after publication, with extensions based on public interest.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.