The LokerDollar.com web up to 85% faster in some cities — by moving our code farther from users.
Sounds backwards. Here's why it worked.
- The edge promise is seductive: run your code in 300+ cities, and it's milliseconds from every user. Newark, Sydney, São Paulo — all local.
- Your code doesn't work alone. It talks to a database. And that database lives in one place. A single page render can mean 6+ sequential trips to it.
- Put your compute "at the edge," and each of those trips crosses an ocean and comes back. The edge didn't remove the distance. It multiplied it. That's how a Sydney request quietly turned into seconds of pure waiting.
The fix was one line of config Cloudflare smart placement stopped stop pinning our code next to the user, and run it next to the database instead. One long trip to reach compute, then all the DB chatter is local and nearly free.
The result, measured on our own traffic (P90 — the slowest 1 in 10 requests):
• Newark ↓ 85%
• Sydney ↓ 82%
• Melbourne ↓ 63%
• São Paulo ↓ 60%
and …12 cities faster overall
The lesson I'm taking with me:
Latency isn't about distance to the user. It's the number of long trips on the critical path. Put compute where it talks the most — for data-heavy apps, that's your database, not the edge.
Top comments (0)