In this interview, Akamai web performance specialists Tim Vera and Robin Marx discuss their approach to achieving tangible performance improvements for modern websites. They define "real" improvements as those impacting actual users in real-world conditions, rather than just in synthetic tests. Key topics include leveraging Early Hints to significantly boost Largest Contentful Paint (LCP), especially for non-cached content, while cautioning against over-optimization that can harm the overall user experience. They explore the inherent trade-off between security and performance, detailing their efforts to minimize the impact of Akamai's own security scripts on Interaction to Next Paint (INP) and empowering customers with granular controls. The conversation also covers network-level enhancements like making Zero-RTT practical by supporting query strings, improving Time to First Byte (TTFB) through DNS upgrades, and innovating in image delivery with advanced chroma subsampling. Looking forward, they express excitement for the potential of the Speculation Rules API and, most notably, Compression Dictionaries, which promise dramatic reductions in file sizes.
Insights from Akamai Experts on Real Web Performance Improvements
00:00:06 Introduction: Tim Vera, a performance specialist at Akamai, explains his role and how he uses his personal scale modeling website as a testbed for new performance features and APIs.
01:09 Getting into Performance: Tim shares that his interest in web performance began when his university website received a good performance rating in a magazine, and later grew from his work optimizing slow enterprise software.
03:04 Introduction: Robin Marx, also a web performance specialist at Akamai, discusses his background, which includes a PhD focused on the real-world performance of HTTP/2 and HTTP/3.
06:45 "Real" Performance Improvements: The term "real" is defined as improvements that benefit actual users and complex, real-world applications, not just single-object tests or on-paper theories.
08:39 Improving LCP with Early Hints: A customer case study demonstrates how implementing Early Hints moved their Largest Contentful Paint (LCP) score from "orange" to "green." The technique is especially beneficial for un-cached, long-tail content where the browser would normally have to wait for the server's full response.
12:14 Best Practices for Early Hints: The experts recommend preloading critical CSS and render-blocking JavaScript and pre-connecting to essential third-party domains. However, they advise against preloading responsive images, as the browser doesn't yet know the correct viewport size to select an image.
17:01 Improving INP and the Security Trade-off: While CDNs have less direct impact on Interaction to Next Paint (INP), Akamai has worked to optimize its own security scripts—which are injected into customer pages—by using modern tools like the Scheduler API to yield to the main thread and reduce their impact.
20:05 Balancing Security and Performance: Akamai provides customers with more granular controls, allowing them to understand the performance-vs-security trade-offs of certain features and choose the balance that best fits their specific needs.
22:35 Making Zero-RTT (0-RTT) Useful: Standard "safe" 0-RTT offers little real-world benefit because it often doesn't work with query strings. Akamai provides a more powerful option that customers can enable for pages with specific query strings, making it effective for use cases like retargeting links.
25:51 Improving Time to First Byte (TTFB): TTFB improvements are the result of many small optimizations, including effective Zero-RTT, TLS 1.3 to origin, and major DNS architecture enhancements, all of which especially help users in the higher latency percentiles (P98, P99).
33:34 Advanced Image Optimization: With a feature called Sharp YUV chroma subsampling, customers can serve visibly higher-quality images for the same file size. This gives them the flexibility to either improve visual quality or apply more aggressive compression while maintaining perceived quality.
39:32 Future: Compression Dictionaries: Robin highlights the game-changing potential of Compression Dictionaries. A proof-of-concept demonstrated a 94% byte savings on a large HTML file, reducing it from 490 KB (compressed) to just 26 KB.
43:30 Contact Information: To get in touch with Tim or Robin for professional inquiries, they recommend connecting with them on LinkedIn.