Turkey is well-connected to DE-CIX Frankfurt. Nigeria has direct fiber to Europe (SAT-3/WACS) but often backhauls to Europe for US routes anyway. So EU-West makes sense for both.
The real outlier is South Africa. Vultr has Johannesburg now, which changes the calculus completely. Before that, EU-West at 160ms beat US-East at 220ms for ZA.
From Chicago, my latency to everywhere:
US-East is obvious for me personally, but I have users in Poland and Germany who complain. I added a $5 Vultr AMS as a read replica for database queries. Write master stays in NYC.
Not perfect but cut EU complaints by 80%.
Member
How do you handle replication lag? I've tried this with PostgreSQL streaming replication, 95ms RTT means 50-100ms lag under load. For my use case (forum) that's fine. For anything real-time it's painful.
Unix philosophy: do one thing well. Your architecture does two things mediocre.
Member
Easy to say from Prague where both regions are 20ms away. My WordPress customers in Montreal don't care about philosophy, they care that their admin panel loads in 2s instead of 5s.
I put them on KnownHost Amsterdam for EU customers, KnownHost Atlanta for US. Same stack, same backups, two bills. Customers don't know the difference.
Member
I buy decommissioned servers from German datacenters and colo in Riga. My latency to EU-West is 25ms, to US-East 120ms. For my scraping workloads it doesn't matter.
What does matter: Hetzner auction prices went up 30% this year. Used to get i7-6700 for €25/mo, now €34. Contabo dedicated looks tempting at €40 for worse CPU but more disk.
Anyone using Contabo dedicated in Nuremberg? Network share same as VPS?
Member OP
Different upstream from what I've seen. Dedicated gets less contention, but the AS-path is the same. I don't have dedicated there though, just VPS.
Back to the original topic: I've uploaded the full CSV. 12 sources, 4 destinations, 100 packets each. Link in my profile if anyone wants to run their own analysis.
Conclusion I'm settling on: US-East origin + Cloudflare for static, accept 200ms+ for APAC until revenue justifies Vultr SIN or similar.
Member
Or use a cheap NAT box like I mentioned. $4 at RackNerd Amsterdam isn't ideal for JP but it's something. Vultr Osaka is $6 for proper VPS with native IPv4.
For "accept 200ms" you could also look at Hostinger in Singapore. Their VPS starts at $4, Singapore datacenter. I haven't tested latency though.
Hostinger VPS is KVM, not containerized like some cheap hosts. I tried their Vilnius location for a month. Network was fine, support was slow. Singapore might be better — https://www.hostinger.com.
Still, for global audience the real answer is Cloudflare Argo or similar. Pay for the smart routing instead of guessing at geography.
Argo helps last-mile, not origin-to-PoP. If your origin is in US-East and user is in Sydney, Argo gets you to Cloudflare's Sydney PoP fast, but then it's still 210ms from PoP to origin for dynamic content.
For API calls that can't be cached, you're stuck with origin latency. Argo Smart Routing is for the eyeballs-to-Cloudflare leg, not the origin leg.
This is why workers at the edge matter for dynamic. But that's a different thread.
Agree but that's architecture, not hosting choice. OP asked about one origin.
I've been running this $15/mo three-region setup for two years. It's overkill for static, necessary for multiplayer game state. For web apps with some dynamic content, I'd still pick one origin and optimize with cache headers.
My actual bill is higher because I also have monitoring and backups in each region. Call it $25/mo real cost.
