Hacker News Reader

Comment by gwright | Hacker News Reader

> ATM was the worst network protocol ever devised

Having 53 byte payloads is something only a committee could love.

ATM cells have 48 byte payloads.

Supposedly that was because different groups involved in the standardisation process wanted either 32 or 64 byte payloads.

That's definitely a "split the baby down the middle" kind of solution.

More like a "fuck all you guys, I'm going home" kind of solution, if you ask me.

I think there's a similar story for the HDTV 16:9 aspect ratio.

Supposedly it was the largest size the French contingent to the standards committee would accept. They wanted a very small sample size to reduce the need for echo cancellation in their network.

conveniently about the size of a typical ip+tcp header

yeah. 53 bytes was the total packet size, not the payload size. But oh hey! lets pick a large prime number for the number of bytes that a payload could be transmitted in. Lets also pick a small prime number of bytes so we can't do byte boundaries.

Needless to say, I am glad that TCP/IP won. Moving on.

First, ATM is OSI Layer 2, where PDUs are not called "packets", they are called "frames". ATM calls these frames "cells", so a cell is 53 octets (which literally means 8 bits, because guess what, not all bytes are that size.) ATM also covers Layer 1 (where PDUs are called "bits" or "symbols") and Layer 3 (where IP starts).

Second, TCP/IP is Layer 3 and up, which means it runs on ATM just fine, like it runs on ISDN, Ethernet, Token Ring, PPP, and the various ITU standards for modems.

ATM is commonly used in DSL implementations, so it also "won" as far as DSL has penetrated as a broadband technology.

At least some telco types absolutely wanted to use ATM for everything. The idea was that instead of the completely anarchy of letting anyone connect to anyone else, you'd use ATM to build up sane, billable virtual circuits between endpoints. Basically, your computer would "call" a server, and Ma Bell would helpfully bill you for the connection and for the amount of data you passed. When you were done, you'd disconnect. I mean, it'd worked for years for voice, so why not data?

Fortunately, that insanity lost.

Um. Is joke? Loads of cloud provider have billing related to the amount of data shipped. Freaking Route53 bills almost per query.

No. They wanted to bill per connection. You know how you use to be charged to call another phone? The telcos wanted that connection-level billing for data.

> Fortunately, that insanity lost.

It lost to ad-supported websites, affiliate links, third-party cookies, tracking and fingerprinting?

Also, ATM was an integral part of SONET/SDH PSTN backbones for decades, so unbeknownst to the consumer, we were using ATM "for everything".

There was talk (I was in the room) of using it, or a successor, as the transport of the Internet.

Imagine replacing IP with ATM, and paying a telco every time you created a virtual circuit for the things we use TCP for today. That was the grand vision.

See https://www.wired.com/1996/10/atm-3/ for the debate at the time.

What you're implying is that transport providers would've charged more than the market can bear, as compared to the flat-fee typical of ISPs in these United States. It's interesting to note that in other parts of the world, such as the ones which used ISDN, Internet usage has been metered, albeit not per-server-connection.

You're also implying that there's something special about the technical design and implementation of ATM that enables monetization based on per-connection events. Well, let me introduce you to TCP, which sets up virtual connections... circuits... at the transport/session layer. There is no technical reason that ISPs (many also happen to be telcos) couldn't charge on a per-TCP-session level in the same way as ATM's SVCs weren't used.

What I'm saying is that the consumers have been monetized anyway, via data mining, engagement, and tracking, and while it's not cash coming out of our pockets, it's still us trading something of value for access to those resources. In fact, the consumer has resources so valuable, that there are whole classes of malware like coinminers, who literally capitalize on them, and there are parallel benign apps which do the same, only with consent.

Perhaps if the telcos had been willing to charge per-VC, our attention and PII would still be monetized, and they'd get us coming and going. Perhaps it's a false dichotomy. I don't know.

For a meteorology focused response I don’t think the use of meteorological summer is too surprising. https://www.ncei.noaa.gov/news/meteorological-versus-astrono...

Still better than 193 bit frames

I assume that’s a T1 dig? It worked quite well for the intended purpose - carrying voice traffic.

So did 53 byte frames, which was due to voice-specific requirements. Could it fit better? Sure. It was after all a compromise so equipment was interoperable.

The 53 byte ATM cell was a compromise for a particular community of voice networks. And the overhead and small cell size never really worked "well" -- there are a bunch of reasons why Ethernet and IP won out, and low average overhead and the ability by the late 90s to scale switching and routing gear up to gigabit speeds and beyond were two big reasons.

ATM collapsed under its own weight.

Reply on news.ycombinator.com