Speed Week: Why "Fast" Needs a Better Definition
The technology industry loves a headline metric. Smartphone makers tout megapixel counts, ISPs sell bandwidth tiers, and web performance vendors push Time to First Byte (TTFB). The problem: none of those numbers, on their own, tells you what you actually care about. A 200-megapixel camera can still take mediocre photos if the sensor and lens are weak. A 1 Gbps connection won't make a sluggish, high-latency webpage load any faster. And a server that returns its first byte quickly can still deliver a page that feels slow to a human being.
Cloudflare's Speed Week 2023 is built around a different question: What does speed look like from the user's perspective, and how do we measure that honestly? The week's announcements span new products, performance benchmarks, and a critical look at the metrics the industry has leaned on for years.
The TTFB Problem
TTFB is the time it takes for a web server to respond to a request with its first byte of data. It's easy to measure, and popular testing tools report it prominently, so it has become a default comparison point. Large enterprises even use software to route traffic between CDNs based on which provider posts the lowest TTFB in a given region.
The trouble is that TTFB measures the network and server, not the user experience. It says nothing about the render-blocking resources that load after that first byte. It ignores the parallel file loading made possible by HTTP/2 and HTTP/3 multiplexing. And it can't account for technologies like Early Hints, Rocket Loader, or HTTP/2 prioritization that are designed to eliminate render-blocking delays. A site with a stellar TTFB can still have a terrible Largest Contentful Paint (LCP) score.
As one industry observer put it, TTFB "is not a user-centric measurement, it's a networking measurement." It tells you how fast a server can produce a response and how quickly that response traverses the network—not how long a visitor actually waits before they can interact with your page.
Real User Monitoring Takes Over
Cloudflare's answer is a wholehearted bet on Real User Monitoring (RUM) as the standard for web performance. RUM tracks the experience of actual visitors, typically through the Core Web Vitals:
- Largest Contentful Paint (LCP): how long it takes to load the largest visible image or text block.
- First Input Delay (FID): the time between a user's click or tap and the browser responding.
- Cumulative Layout Shift (CLS): visual stability, e.g., whether the page jumps under your finger at the last second.
Those metrics reflect what RUM can capture that TTFB cannot: the effect of multiplexing, prioritization features, and other optimizations that change how a full page assembles in the browser. Cloudflare is also preparing for what comes next, including a deep dive on "Interaction to Next Paint" (INP), the emerging Core Web Vital, and what it means for businesses.
The company is introducing a new product that pairs synthetic testing from global vantage points with RUM data, giving administrators a view of how a site performs in the real world. It will be available at every plan level and serves as the new home for application performance on Cloudflare.
Speed Across the Stack
Cloudflare is not limiting this week's benchmarks to websites. The company is publishing updated performance data across its product lines, claiming fastest results in compute, DNS, network, Zero Trust Network Access (ZTNA), Secure Web Gateway (SWG), and object storage.
For Zero Trust, performance is an employee experience issue. If your Secure Web Gateway is in London but your employee is in Johannesburg, every request—Slack messages, Zoom calls, SaaS logins—must travel to the UK and back. That kind of latency is painful and demoralizing. And in a globally distributed remote workforce, these slowdowns can go unnoticed by IT. Cloudflare is unveiling a Zero Trust offering designed to give CIOs visibility into how their employees are experiencing critical business systems, so connection problems get found and fixed quickly rather than lingering in the dark.
Another focus is cache purge. Purge latency is the time it takes to delete an outdated cached file from every server in the CDN. For a business hosting a price list or a time-sensitive document, that number matters—customers could see stale information while old copies are still serving. The week will include a look at Cloudflare's redesigned purge architecture and its resulting performance numbers.
Making the Fast Path Practical
Speed also means removing the barriers that keep customers stuck on slower infrastructure. The Descaler program helps migrate enterprises from Zscaler to Cloudflare in hours. Super Slurper handles moving object storage out of S3. And tools like Turpentine convert legacy VCL configurations. The common thread is that getting onto a faster network shouldn't require a months-long project with a vendor lock-in ransom.
Throughout the week, Cloudflare is covering API acceleration, end-to-end Brotli 11 compression, one-click optimizations that cut page load times by 30%, and more—all framed around the same principle: measure what your users actually experience, not the proxies that are easiest to test.
