Hacker News Reader

Comment by beaconstudios | Hacker News Reader

I think it's interesting for how simple and informal it was in 1995. I always imagined that port allocation was a much more formal process more akin to an ISO or ECMA standardisation, but nope - Tatu sent an email asking for it and Joyce basically just replied with "sure thing, it's yours". For something as critical as port allocation it's pretty interesting/funny how casual it was.

View on HN

The process hasn't really changed much (I applied for a port allocation 5 years or so ago). There's a little questionnaire you fill out, and there might be some back-and-forth over email to clarify the answers but basically it's no more complicated than you email someone and if you fit the criteria they reply with your port allocation.

What has changed is that there is essentially no chance of getting a privileged port allocated now.

Yep it's still very easy and informal. I did it a year or two back to get a DNS-SD entry [1] defined. Just fill out the form [2] and do some back and forth - even updating it is easy. Also yes, a dedicated port is very hard to get - you thought IPv4 was running out ;)

One of the side effects of them being such a widespread and networked group is that when I made my application, they suggested I go talk to one of my coworkers (as loosely defined as that is at Microsoft) since he helped standardize the body anyhow. A bit of zen - "the answers you seek were within you[r org chart] the whole time"

1. https://www.iana.org/assignments/service-names-port-numbers/...

2. https://www.iana.org/form/ports-services

Reminds me a bit of how Eli Lilly landed a class A subnet. They applied for a class B sometime around 1990, but Postel found out how large their company was and gave them an A instead.

(Update: s/network/company/

I doubt they had all that many networked computers at the time, but fuzzy memories of a third-hand story.)

Halliburton (34.0.0.0) and Ford (19.0.0.0) too.

HE sort of should have realized there aren't enough class A:s even for companies of Eli Lilly's size.

IPv4 was an experimental protocol.

http://dltj.org/article/vint-cerf-ip-addressing/

Hah. Just like sockets were a "okay we'll just use this hack until we come up with something better." I think sockets (or what became sockets) crystallized (or started that process) around MULTICS?

He apparently later lamented that Lilly wouldn't return it. I believe they've released large chunks of it by now.

When I registered my first .com address in 1995, I emailed "the guy" who handled them after getting the NCSA webserver up so they could prove I had a server.

Things were considerably more casual in the 90s...

I hope you got a good one!

RFC's were informal. Kind of like 'Sup, here is my proposal for...'

https://en.wikipedia.org/wiki/Request_for_Comments

"Unlike the modern RFCs, many of the early RFCs were actual requests for comments and were titled as such to avoid sounding too declarative and to encourage discussion"

Meanwhile Author's found that very few people were commenting on them, and decided that they might as well declare and standardize from the get go ;)

Critical how?

Its just an informal agreement. Its just a 16bit number.

critical in the sense of avoiding collisions between protocols. Imagine if redis and mysql had ended up with the same default port, and how much of a nightmare that would be.

It would only mean you could not run redis and mysql on the same host, and your default assumption about what port to connect to might be wrong 50% of the time. Easily handled, though.

There are certainly cases of different software using the same unprivileged ports by default. It's probably more rare for the privileged ports.

Who runs redis and mysql on public facing ports?

Whats the big deal if you run ssh on port 443 as many do, together with nginx on the same port?

Or if you run a web-server on port 22 and irc on 21.

>Whats the big deal if you run ssh on port 443 as many do, together with nginx on the same port?

The deal is that if we're talking of TPC apps on a single network interface, you can't.

The deal is also that it's handy to know where you can expect to see a common service running.

The deal is that if we're talking of TPC apps on a single network interface, you can't.

No, but you can fake it with sslh: http://www.rutschle.net/tech/sslh.shtml

Reply on news.ycombinator.com