How We Achieved Sub-50ms Latency Across 6 Continents
The architecture decisions behind our global edge network — from anycast routing to peering strategy.
50 milliseconds. That’s the time it takes light to travel halfway around the world in a fiber optic cable — theoretically. In practice, you lose time at every hop: router processing, BGP convergence, last-mile ISP congestion.
Here’s how we get close to the theoretical limit.
Anycast routing
Every Nodus edge node announces the same IP range via BGP. When a user connects, their request follows the shortest BGP path — which almost always maps to the nearest physical node. No DNS tricks, no geo-IP databases, no JavaScript redirects.
Anycast is simpler, faster, and more reliable than any DNS-based routing we’ve ever tested. The protocol does the right thing by default.
Dense peering
Being close to users isn’t enough. You also need to be close to the networks they use. That’s why we peer directly with 200+ networks worldwide — not just tier-1 transit providers, but regional ISPs, mobile carriers, and content platforms.
In Shanghai, we peer with China Telecom, Unicom, and Mobile. In Cape Town, we’re in Teraco CT1 alongside every major African network. In Los Angeles, we’re in One Wilshire — the most connected building on the West Coast.
Smart failover
When a node goes down — and they do, eventually — anycast automatically reroutes traffic to the next-nearest node. No human intervention, no DNS propagation delay. The BGP withdrawal propagates in seconds.
The result
| Route | Before Nodus | With Nodus |
|---|---|---|
| Shanghai → US East | 280ms | 42ms |
| Cape Town → Europe | 180ms | 58ms |
| Tokyo → London | 220ms | 38ms |
These aren’t best-case numbers. They’re medians, measured over 30 days across real user traffic.
