This is an interesting and fun class exercise but the results were much worse than I expected.
I'm reminded of Carl Sagan's Cosmos where he explains how Eratosthenes calculated the radius of the earth, to a great degree of accuracy, by comparing the shadows of two distant obselisks. I wonder if there is a modern equivalent that would give similar accuracy by observing side-effects of our telecom architecture, or our other very futuristic systems in comparison to Eratosthenes' obelisks.
One thing about traceroute that I doubt most casual users think about (maybe I'm wrong) is that its output only shows you the forward path. That is, it shows you the path your packet takes from when it leaves your machine to when it gets to the destination. The path that a response packet originating at that destination and travelling back to you can take a different path (sometimes wildly different) and - while you'd see that captured in latency - you wouldn't see it captured in the list of routers.
So a packet from Seattle to Hawaii may take the most direct undersea cable, but the response packet may be routed through California, adding significant geographic distance.
This comment is spot on. This is asynchronous nature of the computer networks. While it is easy to control the path within smaller or enterprise networks, it is very likely that the reverse path on the internet is going to be different compared to forwarding path. Following that, the real challenge is, in case that you are dealing with some issue, to detect where the issue is, given that it might be happening on the reverse path that you don't have visibility into.
A similar mistake can easily be made when using iperf to measure throughput or identify performance problems.
In TCP mode the results reflect a combination of forward and return path effects, though the return path usually has less effect on total throughput than forward path. Only with UDP mode can you accurately measure unidirectional path performance.
I used to do traceroute to addresses far away like Japan or Australia to which way it takes. Packets often went through the US even when you'd think there's a shorter route.
These days traceroute to a random .jp or .au just gets you to the nearest CloudFlare or AWS, which is a bit sad in a way.
If there was a shorter route and you took longer one you are dealing with suboptimal routing :)
However, "is a bit sad in a way" part of sentence is interesting one. Edge services hosted within AWS/Cloudflare/Akamai improved customer experience significantly given that waiting for trans-Atlantic or trans-Pacific latencies is not thing any more.
It just shows how much the internet has changed. Now all our online traffic is increasingly being limited to a smaller and smaller set of networks/companies. You get less of a "We're a community of networks all throwing packets around for each other to make the internet happen" feeling and more of a "We're just another source of data to mine by a small number of cloud services." sort of feeling I guess.
Can traceroute be used to prove that the Earth is spherical? Or at least show the statistical probability that it is shaped more like a sphere than a flat surface?
Not sure about definitive proof, but one could trace between a few directly-connected nodes i.e. along the pacific rim. Then, making the same assumptions as the article (i.e. cables lie along great circles, most of the measured time is spent in traversing the cable, etc), demonstrate that those connected nodes do not lie on a flat plane.
The more nodes one has mapped, (i.e. distance measured to surrounding nodes) the less likely one is able to orient those nodes into any shape other than a sphere.
I think yes. If we assume reliable conversion of traceroute results to implied distances, and can generate selected data (routes) of our choice. That is, we can see results of remote hops that aren't directly connected to us. Then, ultimately, the resulting matrix of results would (mathematically, ++) conform to a spheroidal shape and no other.
++ think Map Projections, and the associated math.
It'll add some minor noise but if you suspect it to be at significant factor you can effectively completely eliminate the noise by picking the lowest values over multiple runs. Something like mtr will report jitter statistics for you.
In architecture school, my partner had to map a landscape using old-fashioned triangulation: using measurements between fixed points to build a map of triangles. The problem is, your measurements are usually a bit off (e.g. you measure from different points on a tree's trunk, the tape was sagging, you assume measurements are planar but there is topography, etc. If you aren't careful about the order in which you build your map, errors compound.
We developed some software to optimize for least squared error and then plot points. Not a sophisticated algorithm, but it worked well! https://github.com/cproctor/triangulation
20 comments
[ 2.6 ms ] story [ 59.9 ms ] threadI'm reminded of Carl Sagan's Cosmos where he explains how Eratosthenes calculated the radius of the earth, to a great degree of accuracy, by comparing the shadows of two distant obselisks. I wonder if there is a modern equivalent that would give similar accuracy by observing side-effects of our telecom architecture, or our other very futuristic systems in comparison to Eratosthenes' obelisks.
So a packet from Seattle to Hawaii may take the most direct undersea cable, but the response packet may be routed through California, adding significant geographic distance.
In TCP mode the results reflect a combination of forward and return path effects, though the return path usually has less effect on total throughput than forward path. Only with UDP mode can you accurately measure unidirectional path performance.
These days traceroute to a random .jp or .au just gets you to the nearest CloudFlare or AWS, which is a bit sad in a way.
However, "is a bit sad in a way" part of sentence is interesting one. Edge services hosted within AWS/Cloudflare/Akamai improved customer experience significantly given that waiting for trans-Atlantic or trans-Pacific latencies is not thing any more.
Or, perhaps optimal surveillance [1][2]. I can only assume this sort of thing has expanded substantially since then.
1. https://theintercept.com/2018/06/25/att-internet-nsa-spy-hub...
2. https://tcf.org/content/report/surveillance-without-borders-...
The more nodes one has mapped, (i.e. distance measured to surrounding nodes) the less likely one is able to orient those nodes into any shape other than a sphere.
++ think Map Projections, and the associated math.
(I've used it with street intersection data from Wikidata and it came up with a surprisingly good map of Berlin, only mirrored.)
We developed some software to optimize for least squared error and then plot points. Not a sophisticated algorithm, but it worked well! https://github.com/cproctor/triangulation