Score:1

How is ping so low, lower than theoretical limit?

br flag

According to this article, we would expect a ping of at least 60 milliseconds for 10,000km of distance:

enter image description here

Ireland is roughly 10,000km away from California, so this is the ping we would expect.

I am pinging api.kraken.com from Ireland, which is located somewhere in California.

How is my ping so low - 4 milliseconds?

I am using the ping command in Command Prompt to test this:

enter image description here

And it's definitely not just me getting this low ping. Here is a report from check-host.net:

enter image description here

Here is the location of the server, using iplocation.net:

enter image description here

Here is a traceroute if useful:

enter image description here

cn flag
Ah, no you did not. I did literally not type the quote - I copy/pasted it, and it STILL says 10km. Ah, NOW you changed it. Yeah. NOW it says 10.000km. Which is too big, btw., - the distance is more like 8000, if you want a round number
br flag
@TomTom See edit to comment above.
br flag
@TomTom I rounded it just for simplicity.
Score:10
br flag

For me, traceroute api.kraken.com ends with

 8  ae-2-3602.edge3.Berlin1.Level3.net (4.69.159.5)  15.422 ms  16.278 ms  17.423 ms
 9  unknown.Level3.net (212.162.40.34)  12.069 ms  12.045 ms  9.120 ms
10  104.16.214.191 (104.16.214.191)  9.071 ms  10.039 ms  4.867 ms

From a colo server I rent, it ends with

 7  as13335.dusseldorf.megaport.com (194.146.118.139)  3.027 ms  3.000 ms  2.965 ms
 8  104.16.213.191 (104.16.213.191)  2.419 ms  2.451 ms  2.386 ms

The round-trip times there have a lot of noise on them as we only have three probes per host, but we can be fairly confident that there is no transatlantic link involved.

What is happening here is that Cloudflare are announcing the same IP address block from multiple datacenters that all have copies of the services offered, so each client is connected to the closest one.

Such a setup can fail if routes change around while a connection is active, but generally only few clients are affected and the majority of clients will just rebuild the connection, possibly to a different instance this time.

Keeping the different instances synchronized to simulate a single service is part of Cloudflare's business model.

IP geolocation is guesswork at best, and certainly breaks with such setups since the same IP is used in different places. The ipinfo.io service is likely in San Francisco, so the next instance of api.kraken.com they found is their neighbor.

cn flag
Yes this is essentially CloudFlare's business model. Pushing stuff out to the edge.
br flag
Thanks that is very useful information.
cn flag
There is another important thing here - as you ping kraken API. The shown latency is NOT REAL. As in: it is the latency, but only to cloudflare. Any request and any data you pull will be relayed from this, so while you see a lower ping, the round trip time to the API endpoint is 2x REAL latency, and an order arrives at the API in REAL latency, not the shortened one.
in flag
The only thing to improve on in that answer is to mentioning "anycast IP"
Score:2
cn flag

The api.kraken.com is behind CloudFlare Edges. The IP address is known as anycast IP address. All users will be connected to the nearest local edge server instead of one physical server. In this case, users will enjoy the lowest latency.

I can confirm this by performing traceroute from 3 different continents. Please see the traceroute below and confirm that netrouting to the nearest local CloudFlare edge server.

Traceroute from USA

%> ip2trace -p api.kraken.com -d /usr/share/ip2location/DB4.BIN

IP2Location Geolocation Traceroute (ip2trace) Version 8.0.0
Copyright (c) 2021 IP2Location.com [MIT License]
https://www.ip2location.com/free/traceroute-application

1. 50.97.82.1 0.544 ms ["US", "United States of America", "Texas", "Dallas", "SoftLayer Technologies Inc.]
2. 173.192.118.144 0.398 ms ["US", "United States of America", "Texas", "Dallas", "SoftLayer Technologies Inc.]
3. 50.97.18.192 1.135 ms ["US", "United States of America", "Texas", "Dallas", "SoftLayer Technologies Inc.]
4. 206.223.118.145 1.914 ms ["US", "United States of America", "Texas", "Dallas", "Equinix Inc.]
5. 104.16.212.191 1.533 ms ["US", "United States of America", "California", "San Francisco", "CloudFlare Inc.]

Traceroute from France

%> ip2trace -p api.kraken.com -d /usr/share/ip2location/DB4.BIN

IP2Location Geolocation Traceroute (ip2trace) Version 8.0.0
Copyright (c) 2021 IP2Location.com [MIT License]
https://www.ip2location.com/free/traceroute-application

1. 176.67.168.129 0.511 ms ["FR", "France", "Ile-de-France", "Paris", "UK-2 Limited]
2. 149.6.166.145 0.588 ms ["FR", "France", "Ile-de-France", "Paris", "PSINet Inc.]
3. 154.25.5.125 1.130 ms ["FR", "France", "Ile-de-France", "Paris", "Cogent Communications Inc]
4. 130.117.1.45 1.659 ms ["FR", "France", "Ile-de-France", "Paris", "Cogent Communications Inc]
5. 154.54.60.126 1.382 ms ["FR", "France", "Ile-de-France", "Paris", "Cogent Communications Inc]
6. 149.11.0.126 1.708 ms ["FR", "France", "Ile-de-France", "Paris", "CloudFlare Inc.]
7. 104.16.214.191 1.661 ms ["US", "United States of America", "California", "San Francisco", "CloudFlare Inc.]

Traceroute from Singapore

%> ip2trace -p api.kraken.com -d /usr/share/ip2location/DB4.BIN

IP2Location Geolocation Traceroute (ip2trace) Version 8.0.0
Copyright (c) 2021 IP2Location.com [MIT License]
https://www.ip2location.com/free/traceroute-application

1. 174.133.118.150 0.295 ms ["SG", "Singapore", "Singapore", "Singapore", "SoftLayer Technologies Inc.]
2. 50.97.18.202 0.943 ms ["SG", "Singapore", "Singapore", "Singapore", "SoftLayer Technologies Inc.]
3. 104.16.214.191 0.871 ms ["US", "United States of America", "California", "San Francisco", "CloudFlare Inc.]
mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.