Amsterdam VPS — sub-1ms ping to Cloudflare AMS, NTT routing preferred

adrmnv · May 3

Looking for an Amsterdam VPS with the lowest possible latency to
Cloudflare's Amsterdam edge (138.249.21.0/24, AS13335).

Test target & methodology

Target IP: 138.249.21.1 (or any 138.249.21.x — same edge)
Tool: mtr -rwzc 50 138.249.21.1 from inside the VM
Metric: final-hop best ICMP, plus StDev (jitter matters)
Apples-to-apples: I run the test myself from a real VM, not
just from your Looking Glass. LG-only claims welcome but I'll
verify on a real instance before committing.

Tested so far (final-hop best ICMP to 138.249.21.x)

ASN	Best	Path / Notes
137409	0.53ms	NTT direct peering (LG only — no retail VPS)
57169	0.83ms	Current — Equinix IX (AM1)
20473	1.19ms	AMS-IX (Iron Mountain Haarlem)
22261	1.10ms	ERA-IX (NorthC)
43357	1.30ms	Frys-IX (Iron Mountain Haarlem)
60068	1.60ms	Own backbone routing
56655	1.70ms	Cogent transit (Digital Realty AMS17)
16265	6.30ms	AMS→FRA detour via 3rd-party transit

Best so far: AS57169 at 0.83ms via Equinix IX in AM1.

AS137409's public Looking Glass shows 0.534ms via NTT direct
peering (*.cloudflare.amstnl07.nl.bb.gin.ntt.net), but their
retail VPS isn't sold — only $165+/mo dedicated, currently out of
stock.

What I'm looking for

Any provider matching ANY of these:

NTT (AS2914) direct peering to AS13335 — trace shows
*.cloudflare.amstnl07.nl.bb.gin.ntt.net or similar in hops.
Likely sub-0.7ms.
Direct cross-connect to Cloudflare inside Equinix
AM5/AM6/AM7 (CNI partner). Sub-0.5ms possible.
Anything reproducibly under 0.83ms best ICMP to
138.249.21.x with low StDev (<0.5ms jitter).

Specs needed

1-4 vCPU, 1–4 GB RAM, 20 GB SSD
1 TB/mo traffic, 1 Gbit port enough
Linux (Debian/Ubuntu)
Crypto payment optional but nice
Hourly billing for testing — even better

What I do NOT need

DDoS protection
High bandwidth (10G+)
BGP / own ASN setup
Multiple PoPs / global network
Windows / cPanel / control panels

Budget

$5–50/mo for retail VPS
Up to ~$250/mo for dedicated/colocation if it
reliably delivers sub-0.5ms with verified NTT or
direct CF path

Public Looking Glass appreciated

If your network has a Looking Glass — please post the URL,
I'll run the mtr myself before signing up.

If only private LG, please run from your AMS server:

mtr -rwzc 50 138.249.21.1

…and post the trace (showing your gateway, intermediate
hops, and final-hop best/avg/StDev).

Anti-recommendations (already tested, please don't suggest)

Anything in Iron Mountain Haarlem or Schiphol-Rijk
(~1.2ms+ via AMS-IX, jittery on shared infra)
AS16265 — routes AMS→FRA→back via 3rd-party transit (6.3ms)
BGP-only providers requiring own ASN registration
Iron Mountain–based xTom AMS01

Why so specific

This is for a latency-sensitive client to a Cloudflare-protected
API. Backend-side variance dominates total response time, so
network optimization is marginal — but I'd like to confirm I'm
at the retail-tier floor before accepting it as the structural
ceiling.

If anyone has reproducible numbers beating 0.83ms — would love
to hear from you. If everyone's clustered around 0.8–1.0ms,
that's fine too — I'll know we've hit the realistic floor.

Thanks!

gbzret4d · May 3

0,83 ms is the minimum or average?
Terraswitch rtt min/avg/max/mdev = 0.682/0.714/0.766

Vultr rtt min/avg/max/mdev = 0.720/1.151/1.891
Digitalocean is about the same as vultr

Redswitches rtt min/avg/max/mdev = 0.733/0.752/0.779

Raoul brouns rtt min/avg/max/mdev = 0.496/0.500/0.506

All tested to cloudflare.com so it could also be lower in your case

avsisp · May 3

We are getting from 0.7 - 0.9 to that from our AMS line in Qupra -> Frys-IX (direct CF peering). We also have direct CF peering over Inter-IX which seems to get about the same.

gbzret4d · May 3

https://globalping.io/?measurement=20pmQQ9LEYo103bza00120K7l&by=min

Nekopara · May 3

@as215296

VPSLAYER · May 3

Hello we offer VPS in Amsterdam

Website: https://vpslayer.net
Looking Glass: https://lg-nl.vpslayer.net/

PING 138.249.21.1 (138.249.21.1) 56(84) bytes of data.
64 bytes from 138.249.21.1: icmp_seq=1 ttl=57 time=0.999 ms
64 bytes from 138.249.21.1: icmp_seq=2 ttl=57 time=0.883 ms
64 bytes from 138.249.21.1: icmp_seq=3 ttl=57 time=1.06 ms
64 bytes from 138.249.21.1: icmp_seq=4 ttl=57 time=0.942 ms

--- 138.249.21.1 ping statistics ---
4 packets transmitted, 4 received, 0% packet loss, time 3062ms
rtt min/avg/max/mdev = 0.883/0.971/1.062/0.066 ms

Arirang · May 3

@ShockHosting They are in Equinix AM5 and very reliable.

HOST:                                         Loss%   Snt   Last   Avg  Best  Wrst StDev
  1. AS???    ???                         100.0    50    0.0   0.0   0.0   0.0   0.0
  2. AS???    unn-84-17-33-138.cdn77.com   0.0%    50    0.2   0.3   0.2   0.4   0.0
  3. AS???    cloudflare-ams.cdn77.com     0.0%    50   11.3   5.0   0.5  38.4   7.8
  4. AS13335  141.101.65.1                36.0%    50    1.1   6.0   1.0  36.9   8.2
  5. AS13335  141.101.65.2                 0.0%    50    1.7   3.3   0.7  49.9   7.5
  6. AS13335  138.249.21.1                 0.0%    50    0.8   0.8   0.7   0.9   0.0

Ping

50 packets transmitted, 50 received, 0% packet loss, time 49870ms
rtt min/avg/max/mdev = 0.699/0.758/0.858/0.035 ms

oriend · May 3

What's this used for when sub 1ms required? Some trading? Doubt it's a sneakers bot.

sillycat · May 3

@oriend said:
What's this used for when sub 1ms required? Some trading? Doubt it's a sneakers bot.

Claiming Telegram crypto drops.

Whole /24 is owned by Telegram's Cryptobot entity in Costa Rica.

avsisp · May 3

@sillycat said:

@oriend said:
What's this used for when sub 1ms required? Some trading? Doubt it's a sneakers bot.

Claiming Telegram crypto drops.

Whole /24 is owned by Telegram's Cryptobot entity in Costa Rica.

Ah... now it makes sense. He's trying to beat everyone else to the cryptodrop by running a bot constantly claiming it and hitting it quicker than any human can...

adrmnv · May 3

@gbzret4d said:
0,83 ms is the minimum or average?
Terraswitch rtt min/avg/max/mdev = 0.682/0.714/0.766

Vultr rtt min/avg/max/mdev = 0.720/1.151/1.891
Digitalocean is about the same as vultr

Redswitches rtt min/avg/max/mdev = 0.733/0.752/0.779

Raoul brouns rtt min/avg/max/mdev = 0.496/0.500/0.506

All tested to cloudflare.com so it could also be lower in your case

Solid references, thanks. Few follow-ups so I can compare like-
for-like to my target range (138.249.21.x):

Terraswitch — pulled their LG to 138.249.21.x and got
different numbers per PoP:

PoP	Subnet	min
AMS1	74.118.140.0/21	1.642ms
AMS2	64.130.48.0/20	7.885ms
AMS3	64.130.32.0/20	0.832ms

Your 0.682 lines up closest with AMS3. Which PoP was your test
from — and was it a VPS instance, their LG, or something else?

Vultr — your spread is 0.720 / 1.151 / 1.891. What plan and
region — region-pinned to AMS specifically, or did Vultr place
it elsewhere? Curious whether the wider spread vs. the others
is methodology or actually how their AMS path behaves.

DigitalOcean — which region, and which droplet type? My
intuition is their network tiers might differ but I haven't
tested across them — do you have a sense if it matters here?

Redswitches — what product did you test from (VPS / dedi /
colo)? And do you happen to know which AMS facility or upstream
ASN they peer through? Couldn't find much in their public
material.

Raoul Brouns — is that a service you can actually buy, or
is it @as215296's NLNOG RING node? I couldn't find a
purchasable plan, just a personal/experimental homelab
description, but might be missing something.

Methodology question: were these all from one of your boxes
pinging each provider's IP, or trials spun up on each? The
spread on Vultr looks like single-instance routing variance,
while Redswitches and Raoul are tight enough that they could be
local-VM measurements — want to make sure I'm comparing apples
to apples.

adrmnv · May 3

@Arirang said:
@ShockHosting They are in Equinix AM5 and very reliable.

HOST:                                         Loss%   Snt   Last   Avg  Best  Wrst StDev
  1. AS???    ???                         100.0    50    0.0   0.0   0.0   0.0   0.0
  2. AS???    unn-84-17-33-138.cdn77.com   0.0%    50    0.2   0.3   0.2   0.4   0.0
  3. AS???    cloudflare-ams.cdn77.com     0.0%    50   11.3   5.0   0.5  38.4   7.8
  4. AS13335  141.101.65.1                36.0%    50    1.1   6.0   1.0  36.9   8.2
  5. AS13335  141.101.65.2                 0.0%    50    1.7   3.3   0.7  49.9   7.5
  6. AS13335  138.249.21.1                 0.0%    50    0.8   0.8   0.7   0.9   0.0

Ping

50 packets transmitted, 50 received, 0% packet loss, time 49870ms
rtt min/avg/max/mdev = 0.699/0.758/0.858/0.035 ms

Tried their public LG to 138.249.21.x just now and got:

rtt min/avg/max/mdev = 1.258/1.415/1.532/0.100 ms

That's a 0.55ms gap vs. your trace — wide enough that it's
probably different paths, not jitter. Was your test from an
actual VPS instance, or somewhere else? And which plan/location
specifically — their site lists Amsterdam as one option but
doesn't break out AM5 vs other facilities.

The CDN77 transit hop in your mtr (cloudflare-ams.cdn77.com)
looks like a different upstream than what their LG seems to
use, which might explain it. Want to verify the production
path before signing up — if your VPS reproducibly hits 0.7ms
that's a clear improvement over my baseline; if production
matches the LG number, that's a different story.

adrmnv · May 3

@sillycat said:

@oriend said:
What's this used for when sub 1ms required? Some trading? Doubt it's a sneakers bot.

Claiming Telegram crypto drops.

Whole /24 is owned by Telegram's Cryptobot entity in Costa Rica.

Trading workload, not a claim bot — order routing under jitter, not racing reaction time. Backend variance dominates total response time anyway, so the network is marginal.

adrmnv · May 3

@avsisp said:
We are getting from 0.7 - 0.9 to that from our AMS line in Qupra -> Frys-IX (direct CF peering). We also have direct CF peering over Inter-IX which seems to get about the same.

Picked up your smallest AMS VM yesterday to verify. Path checks out — mtr from inside the VM goes gateway → AS13335 directly at hop 2, no intermediate IX, so the direct peering is real on this subnet:

hop	host	best
1	_gateway (10.0.0.1)	0.2ms
2	AS13335 141.101.65.1	1.4ms
3	AS13335 141.101.65.97	1.1ms
4	AS13335 138.249.21.3	1.1ms

100-ping result: min/avg/max/mdev = 1.011/1.186/1.522/0.081 ms

Solid trace — direct peering as advertised. Just lands ~0.18ms
slower than my baseline (~0.83ms) on this specific edge,
which I'm guessing is a Qupra-vs-Equinix-AM1 physical-distance
effect rather than anything routing-related. For my sub-ms
hunt that's enough that still wins, but worth noting your
direct peering claim verifies cleanly on real VM testing.

gbzret4d · May 4

@adrmnv said:

@gbzret4d said:
0,83 ms is the minimum or average?
Terraswitch rtt min/avg/max/mdev = 0.682/0.714/0.766

Vultr rtt min/avg/max/mdev = 0.720/1.151/1.891
Digitalocean is about the same as vultr

Redswitches rtt min/avg/max/mdev = 0.733/0.752/0.779

Raoul brouns rtt min/avg/max/mdev = 0.496/0.500/0.506

All tested to cloudflare.com so it could also be lower in your case

Solid references, thanks. Few follow-ups so I can compare like-
for-like to my target range (138.249.21.x):

Terraswitch — pulled their LG to 138.249.21.x and got
different numbers per PoP:

| PoP | Subnet | min |
|-------|---------------------|----------|
| AMS1 | 74.118.140.0/21 | 1.642ms |
| AMS2 | 64.130.48.0/20 | 7.885ms |
| AMS3 | 64.130.32.0/20 | 0.832ms |

Your 0.682 lines up closest with AMS3. Which PoP was your test
from — and was it a VPS instance, their LG, or something else?

Vultr — your spread is 0.720 / 1.151 / 1.891. What plan and
region — region-pinned to AMS specifically, or did Vultr place
it elsewhere? Curious whether the wider spread vs. the others
is methodology or actually how their AMS path behaves.

DigitalOcean — which region, and which droplet type? My
intuition is their network tiers might differ but I haven't
tested across them — do you have a sense if it matters here?

Redswitches — what product did you test from (VPS / dedi /
colo)? And do you happen to know which AMS facility or upstream
ASN they peer through? Couldn't find much in their public
material.

Raoul Brouns — is that a service you can actually buy, or
is it @as215296's NLNOG RING node? I couldn't find a
purchasable plan, just a personal/experimental homelab
description, but might be missing something.

Methodology question: were these all from one of your boxes
pinging each provider's IP, or trials spun up on each? The
spread on Vultr looks like single-instance routing variance,
while Redswitches and Raoul are tight enough that they could be
local-VM measurements — want to make sure I'm comparing apples
to apples.

I've tested it with globalping:
https://globalping.io/?measurement=20pmQQ9LEYo103bza00120K7l&by=min

gbzret4d · May 4

rtt min/avg/max/mdev = 0.239/0.247/0.256 to 138.249.21.1 @adrmnv

Howdy, Stranger!

Categories

In this Discussion

Amsterdam VPS — sub-1ms ping to Cloudflare AMS, NTT routing preferred

Test target & methodology

Tested so far (final-hop best ICMP to 138.249.21.x)

What I'm looking for

Specs needed

What I do NOT need

Budget

Public Looking Glass appreciated

Anti-recommendations (already tested, please don't suggest)

Why so specific

Comments

Howdy, Stranger!

Quick Links

Categories

In this Discussion

Amsterdam VPS — sub-1ms ping to Cloudflare AMS, NTT routing preferred

Test target & methodology

Tested so far (final-hop best ICMP to 138.249.21.x)

What I'm looking for

Specs needed

What I do NOT need

Budget

Public Looking Glass appreciated

Anti-recommendations (already tested, please don't suggest)

Why so specific

Comments