Oops! Something went sideways.

Looks like the styling got goofed up. Sorry about that, unless it's what you wanted. If this isn't what you were looking for, try force refreshing your page. You can do that by pressing Shift + F5, or holding Shift and clicking on the "reload" icon. (It's the weird circle arrow thing "⟳" just above this page, usually next to where it says https://blog.unitedheroes.net...)

isn't quite ashamed enough to present

jr conlin's ink stained banana

:: The Internet Hates Long Lived Things

First off, this is not about ageism. i’m talking about long lived connections. There are a few folk out there that believe that you can hold a connection between two devices open forever. This is not the case. There are a lot of reasons that a great many things will actively fight your long lived connection. So, here are a few insights from someone who has dealt with Very Long Connections in Webpush and was once naive like you.

Why does the internet hate long lived connections?

Short answer: Money.

Longer answer:
The internet is not free.

Everything about the internet costs money, because everything requires either power or devices. Devices are way more costly because you not only need to buy and power them, you need to shelter, maintain, inspect, and eventually replace them. This includes everything from colocation farms to servers to cables to the conduits that carry the cables and the folks who’s jobs it is to do all that sheltering, maintaining and inspection. The costs may be near infinitesimal for a 10 byte ping, but they’re there, and they add up surprisingly fast.

i’ll also add in that connections between devices also have a software cost. Turns out, there are a limited number of connections that a given computer can accept. There are also constraints depending on the language you use, how much memory you have installed, how fast your CPU is, and how many files you need to have open. There are fun ways to tweak that number and get really high counts, but if you’re doing any actual work with them, you’re going to hit that upper limit. If you’re doing real, serious work (like running TLS so things are secure) boy golly are you going to hit that number and it’s not going to be anywhere near that 10 million connection number someone built for Erlang.

So, in that sort of world where having connections that are basically doing nothing but tying up resources, connections are not going to stick around. You may not want to pay for them, and neither do any of the dozens of intermediary companies what want to maximize profits. They’ll spot a connection as being underused and will simply drop it, since there is probably some other company that wants to use it and send lots of capital producing data over it.

There are tons of reasons a connection could be killed at any time and a whole lot of incentive to ignore any requests you might make to keep a low bandwidth connection up. This includes various “Keep Alive” packets helpfully provided by protocol authors. Those tend to be very light weight dedicated Ping/Ack packets that are sent on a regular cycle. They’re useful if you’ve got a lull for a few minutes, but anything longer than that and the connection is toast. You’re better off crafting a NoOp type message that you fire off regularly. Granted, i fully expect that those will be dropped in the future too once providers use stuff like packet inspection machine learning to further reduce costs and free up “idle” connections.

Well, what about using stateless UDP instead of stateful TCP?

It’s not a bad idea, really. It’s the reason that QUIC is the base for HTTP 3.0, and it’s very clever about making sure that packets get handled correctly. Packets are assigned Server Ids, and cryptography is isolated so data corruption doesn’t cause blockages. Even though, if there’s a connection severance, it’s still dependent on the Client getting back to the Server. The server needs to be at a known, fixed address. That’s neato for things like HTTP, but less so for things like WebPush where the client could be waiting hours or days for a response, and unless the client is actively monitoring the connection (remember, built in KeepAlive packets ain’t enough), it’s basically doing long polling, so you’re kind of back at square one.

(There’s definitely something to be said about that for things like WebPush. WebPush’s “Immediate receipt” requirement, like relativistic travel, depends a great deal on the perspective of the parties involved. That’s a topic for another post.)

So, be mindful young protocol developer/designer. The internet is out to get your long lived connection dream and will dance on it’s grave at every opportunity.

Blogs of note
personal Christopher Conlin USMC Henriette's Herbal Blog My Mastodon musings Where have all the good blogs gone?
geek ultramookie

Powered by WordPress
Hosted on Dreamhost.