Revisiting Browser Performance Benchmarking From an Architectural Perspective

Yongye Zhu, Shijia Wei, Mohit Tiwari

Abstract: Web browser are traditionally benchmarked using user-centric page-load times. We found such metrics inaccurate from an architectural perspective, and limit hardware and systems optimization efforts, for performance and energy efficiency. In this article, we introduce <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SysPLT</i> , an accurate and stable metric targeting system optimization for browsers. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SysPLT</i> tracks page load time based on browser's architectural activities. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SysPLT</i> captures on average 2.29× micro-architectural events compared to existing user-centric metrics, and provides insights to correct counter-intuitive system design choices recommended by user-centric metrics. We also show that <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SysPLT</i> is more stable across repeated runs and different machine settings.