macOS command-line tools you might not know about

Neat, I was not aware of `networkQuality`. A good replacement for opening up Speedtest or whatnot when you just want to figure out if the network is slow or something else is up.

It seems to grossly underestimate my upload bandwidth!

same

What are you, a station wagon full of thumb drives?

I can't find any (official) documentation on how it measures the speed. Against which target?

Edit: Found https://www.macinstruct.com/tutorials/how-to-check-your-macs... which says:

"The networkquality tool uses Apple’s CDN (https://mensura.cdn-apple.com/api/v1/gm/config) as a target"

The contents of this file (for me):

  { "version": 1,
    "test_endpoint": "sesto4-edge-bx-021.aaplimg.com",
    "urls": {
        "small_https_download_url": "https://mensura.cdn-apple.com/api/v1/gm/small",
        "large_https_download_url": "https://mensura.cdn-apple.com/api/v1/gm/large",
        "https_upload_url": "https://mensura.cdn-apple.com/api/v1/gm/slurp",
        "small_download_url": "https://mensura.cdn-apple.com/api/v1/gm/small",
        "large_download_url": "https://mensura.cdn-apple.com/api/v1/gm/large",
        "upload_url": "https://mensura.cdn-apple.com/api/v1/gm/slurp"
     }
  }

I find solace in the fact that even at large companies, there's still enough whimsy to name your upload endpoint slurp.

There is also a CLI client for speedtest.net[1] and fast.com[2]! Not included with MacOS, of course, but nice to have around.

[1] https://www.speedtest.net/apps/cli

[2] https://github.com/sindresorhus/fast-cli

It tests your bandwidth while fully saturating your upload AND your download. If maxing out your upload has a huge impact on your download (like, a 70% drop), then that’s probably a sign you could tweak your network for better flow control.

That’s sort of the idea of `networkQuality`. It’s a new idea for how to measure that’s different from the standard speed test.

Through if you run it with the `-s` flag it’ll test them sequentially.

It’s a new proposed standard that’s gaining support. People are building alternative party implementations for it: https://github.com/network-quality/goresponsiveness

Try running it with `-s` to see if you get speed tests that resemble what you expect. The idea is that maxing out both links at once is a better measure of network quality than sequence speed tests. Also it’s new RTT metric is key.

In my case, it appears that maximizing my download has a huge impact on my upload as well, which makes sense from a TCP/IP point of view. In this case, do you mean the ISP's flow control?

I'm sure others know waaay more about this, but I think it's a thing you can improve locally. Though usually I think of it as the other way around, where maxing out upload severely impacts download. My understanding is that this causes bufferbloat[0], making packets queue up for a long time on your gateway, ultimately limiting you to way less bandwidth then you should be able to get.

My one experience with this is on Ubiquiti hardware where there's a feature called "Smart Queues" you can enable. Really it's FQ_CODEL[1] under the hood. If you tell it your real maximum up/down bandwidth, minus ~5%, it'll enforce those limits in a way that prevents buffer bloat and lets you use nearly your full download bandwidth even when your upload bandwidth is maxed out. On Ubiquiti gear this has a CPU impact since it has move some traffic handling from dedicated hardware to the CPU. But it was a huge night and day difference for me. After enabling this, having a couple people on Zoom calls (highish upload) no longer tanked everyone else's download speed.

Also I think this stuff matters more when you have a large multi-user network. For normal home life, definitely not worth it. (In my case it was wifi for ~20 people).

[0] https://en.wikipedia.org/wiki/Bufferbloat [1] https://en.wikipedia.org/wiki/CoDel