WebOptProfiler: Providing performance clarity for Mobile Webpage Optimizations
WebOptProfiler: Providing performance clarity for Mobile Webpage Optimizations
复制标题
DOI：
10.1145/3446382.3449073
复制
发表时间：
2021-02
期刊：
影响因子：
--
通讯作者：
Ghulam Murtaza;Theophilus A. Benson
中科院分区：
文献类型：
--
作者：
Ghulam Murtaza;Theophilus A. Benson
尽管对移动网页优化进行了数十年的研究，但人们对这些优化如何互操作知之甚少。此外，很少有系统的工作来理解这些优化组合的效果。如果不全面了解这些优化是如何相互组合的，以及它们在什么条件下表现出色，操作人员就无法确定采用哪种优化，同样，开发人员也不知道应该把精力集中在哪里。在本文中，我们认为应该要求开发人员评估和描述他们提出的优化和其他优化之间更广泛的交互——这是除了展示他们的方法的潜在好处之外。为了帮助开发人员描述这些更广泛的交互，我们提出了一个分析模型，该模型将web优化分解为虚拟加速函数，这些函数在易于理解的浏览器处理阶段（例如，处理，渲染，布局等）上运行，并且我们提出了一个面向web浏览器的因果分析器，该分析器通过使用其分析模型在页面加载期间加速浏览器的不同部分来经验地探索优化之间的交互。我们的系统，WebOptProfiler，识别并解决了将因果分析扩展到网页优化领域的实际问题，并提供了一种从现成的浏览器跟踪中提取分析模型的算法。
Despite decades of research on mobile webpage optimizations, little is known about how these optimizations interoperate. Moreover, there has been little systematic work to understand the scenarios wherein combinations of these optimizations excel. Without a comprehensive understanding of how these optimizations compose with each other and under what conditions they excel, operators cannot determine which optimizations to adopt, and, similarly, developers do not know where to focus their efforts. In this paper, we argue that developers should be required to evaluate and characterize the broader interactions between their proposed optimizations and other optimizations - this is in addition to demonstrating the potential benefits of their approach. To aide developers in characterizing these broader interactions, we propose an analytical model which decomposes web optimizations into virtual speedup functions that operate on well-understood browser processing phases (e.g., processing, rendering, layout, etc., for an object) and we present a web browser-oriented causal profiler which empirically explores interactions between optimizations by using their analytical models to speed up different parts of the Browser during a page load. Our system, WebOptProfiler, identifies and addresses practical issues in extending causal profiling to the webpage optimization domain and provides an algorithm for extracting an analytical model from readily available browser traces.