Cloudflare reduced the per-entry footprint of a cache holding more than 250 billion DNS entries from 953 bytes to 420 bytes. Replacing growable collections, packing flags, removing repeated owners, and storing record data contiguously freed roughly 100 terabytes across the fleet. Insert throughput rose 43% and lookup latency fell 19%, showing how fewer allocations and better locality can improve both capacity and speed.
Chrome and Firefox are moving toward two-week major release cycles, which means browser-version mix can shift real-user monitoring without an application deploy. Smaller cohorts, more boundaries inside a 28-day window, and long-lived managed releases can all move a percentile. One navigation study also found 7,131,737 observations with unattributed overhead beside only 166 visible redirects, reinforcing the need to name missing time without pretending the remainder is already diagnosed.
A small static site's CDN experiment supplies the break-even check. Warm hits reached 0.239 seconds, cold misses cost 0.366 seconds, and the uncached baseline was 0.281 seconds; caching therefore needed a 67% hit rate merely to break even. With nine daily visitors spread across 77 pages and many edge locations, the real hit rate approached zero and a crawler's slow-page count rose from 38 to 75.
