Cloudflare Releases BEACON Dataset for Unprecedented Web Performance Insights
Cloudflare has launched the BEACON dataset, offering billions of real-world performance measurements from 10,000 major websites to provide a granular view of user web experiences.
Cloudflare has unveiled the BEACON dataset, a massive collection of billions of real-world performance measurements gathered from 10,000 of the world's largest websites. This initiative aims to bridge the gap between the optimized development environments of tech professionals and the often less-than-ideal conditions faced by everyday internet users. By providing objective data on how the web performs across diverse devices, browsers, and geographic locations, BEACON seeks to foster a more accessible and faster internet for everyone.
The BEACON dataset, now available via Google BigQuery, adheres to standards set by the community-led RUM Archive. It includes comprehensive data on Core Web Vitals—Largest Contentful Paint (LCP), Cumulative Layout Shift (CLS), and Interaction to Next Paint (INP)—across various user contexts. The dataset's granularity, offering full histograms rather than simple averages, allows researchers to analyze performance down to specific percentiles, revealing performance disparities that might otherwise be obscured.
Analysis of the BEACON data highlights significant performance differences. For instance, while WebKit-based browsers generally perform well, they exhibit notably slower LCP and INP in 46 countries where they represent a substantial portion of traffic. In Cambodia, WebKit's LCP is reported to be 50% worse than that of Blink-based browsers. Furthermore, the dataset categorizes websites by industry, revealing that government, health, and 'safe for kids' sites tend to perform better than those in the ads, religion, and weather sectors.
BEACON also delves into the sub-parts of LCP and INP to pinpoint performance bottlenecks. For LCP, it breaks down the process into Document TTFB, Load Delay, Load Duration, and Render Delay. The data suggests that for pages exceeding the 'Good' threshold, the primary optimization opportunities lie not in downloading the content itself, but in discovering the LCP candidate and resolving rendering blocks. This challenges the common assumption that bandwidth is the main culprit for slow loading times.
Similarly, INP is dissected into Input Delay, Processing Time, and Presentation Delay. The findings indicate that while JavaScript execution time is a major factor in slow interactions, complex CSS layout recalculations also contribute significantly to presentation delays. Cloudflare offers tools like Zaraz to help mitigate the performance impact of third-party JavaScript, a common contributor to these issues.
The dataset's inclusion of data from the Soft Navigations API is particularly relevant for modern single-page applications built with frameworks like React, Vue, and Angular. This API provides more accurate LCP reporting for client-side navigations, a common pattern in these applications, thus offering a more realistic performance picture for a growing segment of web development.
By making this extensive dataset publicly available, Cloudflare aims to empower developers, researchers, and website owners with the insights needed to identify and address performance issues. Understanding these real-user metrics is crucial for making informed technology choices and prioritizing improvements that benefit the broadest possible audience, ultimately contributing to a faster and more equitable web experience.