[jlmcguire]

I wish the article had mentioned more about the tools they use to track performance of links. My guess is a lot of perfsonar https://www.perfsonar.net/ or iperf nodes. Network performance testing is tough. It's interesting to hear of a company doing it

[quentusrex]

That's actually a pretty close guess. In order to get to the measurement capabilities that we needed, we had to build our own internal systems. Some of those systems had a small mention in this public talk: https://www.youtube.com/watch?v=kyu_wDcO0S4&t=7140s

There are several major issues when trying to measure network paths in real time, but one in particular is how to avoid the assumption about path latency related to RTT/2. The latency of a path one direction rarely is half of the over all round trip latency, usually there are a combination of things that cause path asymmetry.

Anything you or others would like to know more about in terms of network performance measurements?

[jlmcguire]

Thanks for the reply! Nice talk! I spent around a month going down the perfsonar rabbit hole and set some cool stuff up. It was really instructive. At the end of the day work didn't operate a large enough network for us to get great insights from the perfsonar network.