Skip to content
Register

US East vs EU West for a global audience — actual latency numbers

Networking by LichunLars 34 replies 4.2K views
GeorgeNmp

Member

AS64512

#11
LichunLars said:
So EU-West advantage holds across most of APAC except maybe Japan where it's closer to a wash.

I already covered this. The 12ms delta is peering location, not "Cogent bad." If your path exits Cogent at 15454 (Equinix NY5) versus 3356 at 555 (Equinix NY2), that's a 0.5ms walk across the campus, not 12ms. The 12ms is where Cogent hands off in London versus where L3 hands off in New York.

Run a traceroute and check the penultimate hop before transatlantic. If Cogent lands you at LHR first, that's your extra 12ms right there.

iBGP, eBGP, don't care, just peer
hankels

Member

52 VPS and counting

3 #12
GeorgeNmp said:
If Cogent lands you at LHR first

This is exactly what I see on my Contabo NYC box. Cogent path to EU goes: NYC -> LHR -> AMS. Level3 path goes NYC -> AMS direct. The LHR detour adds 11-14ms depending on time of day.

I don't blame Cogent, it's just how their network is built. For $5/mo I'm not complaining.

seedbox, NAS, tape, and three offsite
FlowSana

Member

SRE

#13
hankels said:
For $5/mo I'm not complaining

This is the real takeaway. The latency differences we're debating are often smaller than the variance you get from budget network congestion. Contabo's $5 plan is on a shared 200Mbit port. At 8pm EST I've seen iperf drop to 40Mbit. That matters more than 12ms for most workloads.

If you're running latency-sensitive stuff, pay for dedicated bandwidth or run your own congestion control testing. BBR helps but doesn't fix an oversubscribed uplink.

It's always DNS. Always.
kate3

Member

#14
FlowSana said:
Contabo's $5 plan is on a shared 200Mbit port

Closer to shared gigabit on newer nodes, but the contention is real. I have four Contabo VPS across Nuremberg and NYC. The NYC one hits 900+ Mbit at 3am, drops to 200-400 at peak. Nuremberg is more consistent, maybe because fewer US customers.

Ask me about overselling. The answer is: everyone does it, Contabo just does it more honestly by pricing lower.

virsh list --all | wc -l: 47
kenji3

Member

#15
kate3 said:
Contabo just does it more honestly by pricing lower

Honest overselling is still overselling. I moved my JP site from OVHcloud Beauharnois to Vultr Osaka last year. Price doubled, latency to Japan dropped from 180ms to 8ms. Worth it for my use case.

For global audience without CDN, I still think multi-region is unavoidable. The "US-East vs EU-West" framing is a false dichotomy.

conbini > datacenter snacks
MA

Member

1 #16

From Buenos Aires, my numbers:

  • Contabo NYC: 128ms
  • Vultr AMS: 205ms
  • Hetzner HEL: 218ms
  • OVHcloud Beauharnois: 142ms

US-East wins for South America, no contest. The path to EU-West often routes through MIA first anyway, so you eat the US-East latency plus transatlantic.

I run a small game server for local friends. 128ms is playable. 205ms is not.

quinnbizz

Member

6 #17
march13 said:
128ms is playable. 205ms is not.

NZ agrees. From Wellington:

  • US-East: 210ms
  • EU-West: 280ms
  • Vultr SYD: 35ms

I don't even bother with US or EU for anything interactive. For static sites, Cloudflare handles it. For dynamic, I need local presence.

The whole "pick one region" discussion feels very Northern Hemisphere.

...
LichunLars

Member OP

#18
quinnbizz said:
The whole "pick one region" discussion feels very Northern Hemisphere.

Fair point. My original question was specifically about US-East vs EU-West for a global audience, assuming you had to pick one. The answer from the data is US-East for lower global average, but with huge caveats for APAC and Oceania.

I should have been clearer: this is for a bootstrapped project that can't afford multi-region yet. CDN for static, one origin for dynamic.

1ms or I don't want it
adadude

Member

Script Everything

#19
LichunLars said:
One origin for dynamic

Then your user distribution matters more than any global average. I wrote a script that pulls Cloudflare Analytics and weights latency by country. For my side project:

  • US-East weighted: 94ms
  • EU-West weighted: 87ms

EU-West wins because 35% of users are in India, Turkey, and Nigeria. US-East would win if my audience were Americas-heavy.

Script is ugly but works. Want me to clean it up and share?

minh1987

Member

#20
adadude said:
35% of users are in India, Turkey, and Nigeria

India routes better to EU-West? My testing showed EU-West to MUM was 142ms, US-East to MUM was 187ms. Confirms your numbers.

But Turkey and Nigeria surprise me. I would have guessed US-East for both.

phở at 3AM, deploy at 4

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft