Characterizing GPU-accelerated Web Applications in Browsers

Driven by the rapid advancement of Web immersive experiences, the demand for GPU acceleration in Web applications has become an essential component of the Web ecosystem. However, little is known about the current practices of GPU-accelerated Web applications. To bridge the knowledge gap, we conduct the first comprehensive empirical study of real-world Web applications that leverage the Web Graphics Library (WebGL) for GPU acceleration, i.e., WebGL applications. We construct a dataset consisting of 5,954 real-world WebGL applications and develop a measurement tool to evaluate how these applications perform across various devices, operating systems, and browsers. Our findings reveal the performance defects in these applications, along with the root causes stemming from undesired development practices. Based on the findings, we propose actionable strategies for developers and browser vendors to mitigate the performance defects.