HN Reader

> and a complete 15-second connectivity dropout every few minutes?

> I can’t seem to emulate the connection dropout, but the 10% packet loss should be harsh enough. ... Still, it’s Bad Enough.)

The total dropout is surprisingly bad, probably because it kills a lot of "are we alive?" checks... then when things come back, they all hammer the crappy link and everything just dies in a way that all too often doesn't recover.

Seriously, more developers need to eat their own dogfood on the subway, or in the back of a grocery store that's surprisingly well RF-shielded, or in a modern steel-reinforced-concrete building but without WiFi (so you're on cellular data which can't quite make it through the building skin), or....

> Discord will sometimes just kick you to a loading screen if it thinks your internet isn’t working. If it doesn’t do that, it works pretty okay.

On Android, falling back to EDGE/EGPRS makes Discord just not work at all. I get it - downloading profile pictures and media is something you cannot realistically expect, but being able to receive notifications and text content on currently selected channel is a problem that should have been solved way before $current_year.

This reminds my of 'chaos engineering' and the approach that Shopify took with 'Toxiproxy.' Toxyproxy is this elegant system that provides a REST API for spawn proxy servers. The proxy instances can have 'toxics' added to them that change the connections properties. E.g. you can add latency, packet drops, even experiment with slicing up packets. It allows for something like writing deterministic tests that test whether your software works under many different network conditions. I think its honestly a really smart design.

I'm not sure if the author will be reading this, but I was just curious after reading it's pronoun page, it says it prefers to refer to itself in the third person inanimate, but uses first person pronouns in this article. Are the pronoun choices new, or was there some other choice made specifically for this article?

Not only should developers design for slow internet, they should design for no internet. Shockingly, many parts of the US don't have solid cell phone coverage. Even more cell phone towers have broken uplinks or broken configurations; your phone connects to the cell, but no data flows. Cells get overloaded - at fun things like sports events and concerts. They also get overloaded at very not-fun events like natural disasters where everyone is standing around the (very overloaded) Cell On Wheels tower, trying to get a message out to their loved ones that they're OK or need help, or trying to reach their insurance company, or access a government website to get info about disaster relief, etc.

It's infuriating how many smartphone apps will show you a screen full of the information you need while online, and then when they lose internet connectivity due to a dead spot in cell coverage, yank it away leaving you staring at a blank screen with a spinning progress indicator.

It's also infuriating how so many apps are completely, totally non-functional if they don't have internet, if only because they're not actually a native smartphone app, but a website being run in gussied up web browser instance.

On macOS one can use the traffic shaper dummynet(4), controlled with the dnctl(8) command, which has a nice man page that describes it.

It's always surprising to see how apps degrade under high latency connections and is a good way to test how users will actually experience things in less-than-ideal conditions.

On macOS, there is (was?) "Network Link Conditioner", used to be part of XCode.

I often experience this sort of connection in Germany, either because of our outdated infrastructure or because my data plan ran out and I’m on the almost-no-internet speed.

I test for these things. Developer Tools lets you see how your website performs on such connections, and how long it takes to get a usable page.

On windows, clumsy[1] is a great tool for this. The animated gif on their homepage is a super intuitive demo of the functionality too.

1: https://jagt.github.io/clumsy/

I unintentionally did this recently on the hardware side by messing up crimping my internet cable and not pushing it all the way. Some things like Teams were surprisingly resilient even with 8% packet loss, probably due to its realtime nature

Sounds useful for some tests that I have to do this week for a bug at $work. And looks like `tc` works with Docker too, to avoid messing with my local wnv, e.g. https://medium.com/@kazushi/simulate-high-latency-network-us...

From the article:

> If you want to test services running on localhost, replace enp3s0 with lo. Though some other programs on your system may Not Like It if localhost is slow, so do that at your own risk :3 (Be sure to replace enp3s0 with lo on the undo script!)

In reality, lo will hate to be slow, esp. any application working through local network sockets (cough X11 cough). I experienced this in the olden times of Fedora 4. Somehow I (or a bug) managed to route local network sockets through my eth0. The more load the card has, the slower the X11 applications were. Esp. during launch.

I had to add another card and use it for external traffic, leaving eth0 free.

Is there a html standard thingy to tell the browser "This resource is optional, don't load it if the connection is already slow"? I'm thinking of optional fonts, JS or maybe changing the quality of images. You could probably do this in JS, but it'd be pretty helpful if there was a standard for this

My mom used to be on rural dialup, with 30Kbps on a good day. Back then (200x) you could access the internet that way. Facebook, Gmail, you name it (Gmail definitely worked better in its basic HTML mode ... RIP). It did take patience but everything worked.

Then things got more modern, and she's on a cellular data plan. Once we ran it out on purpose near the end of the month, to see what the 128Kbps fallback is like. 4x as fast as she used to have, good enough, right?

Nope. Half the web didn't load at all, due to timeouts loading resources and such.

And now people are complaining about falling back to super slow 3G.

But with everything bloated 100x since the 200x's, everything has gotten 100x better, right?

I have a fairly basic, though recent, Android phone (Moto G Play 2023). I use an app called Flashfood. It was crisply performant. Then an update got pushed that made mainly cosmetic changes... and slowed it down by a factor of 10-20. Now it feels glacial on the same phone. Do they care? Maybe they should. The super recent, $1K smartphone crowd may not be exactly the Flashfood crowd. My usage of the app is way down now, because it's just not worth the aggravation.

Impressive that this same person has 2 front page HN articles in one day. I don't think I've seen that before.

netem is especially useful when paired with network namespaces, that way you can selectively apply different profiles to different process trees.