Teams often spend weeks optimizing JavaScript while their TTFB sits at 800ms. Core Web Vitals start at the server.
LCP: the 40% you don’t control in the browser
Largest Contentful Paint budgets roughly break down as TTFB + resource load + render. If TTFB is 800ms, no amount of image optimization gets you under 2.5s reliably. Server-side levers:
- NVMe storage — random read latency drops an order of magnitude versus SATA SSDs.
- LiteSpeed + LSCache — cached TTFB under 100ms is normal on our shared platform.
- HTTP/3 — eliminates head-of-line blocking on lossy mobile networks.
- Edge caching — serve HTML from a node near the visitor.
INP: keep the main thread free
Interaction to Next Paint is mostly a frontend concern, but server-timing headers help you prove where the time goes. Enable them and separate backend latency from script cost before refactoring.
CLS: reserve space, always
Layout shift is a discipline problem: explicit width/height on every image and embed, font-display: swap with size-adjusted fallbacks, and never injecting banners above content. (This site ships with CLS regression tests — we practice what we preach.)
Measure from where your users are
Test from the region your customers live in. Our datacenters page lets you download test files from each region to measure real-world throughput from your location.