Hacker News Reader

Comment by brk | Hacker News Reader

Having worked in the telco space, I'm not sure there were ever truly "bellheads" and "netheads". It was mostly an industry of carriers trying to charge as much as possible, while delivering as little as possible. As on example, CallerID cost almost nothing to offer, but rather than just make a free convenience service, telcos charged ~$10/mo. for you to be able to see data that was already present anyway.

As internet connectivity became a standard utility for most households, telcos were forced to offer it, or die. Voice-specific equipment then became more of a liability than an asset, so we saw the networks switch to being primarily IP carriers, and voice calls got degraded to using things like SIP out of convenience for the carriers.

View on HN

> I'm not sure there were ever truly "bellheads" and "netheads".

In my experience, I've seen these terms used to refer to circuit-switched/deterministic vs packet-switched/opportunistic routing.

Bellheads, having done frequency-division and then time-division multiplexing, derisively referred to packetized voice as "statistical multiplexing" and ridiculed its pathetic jitter characteristics. (Radio stations still use ISDN for some things because it guarantees latency and jitter performance that Ethernet can't.)

Netheads, convinced that when all you have is a hammer, everything looks like a packet, said bellheads and their obsessions with latency and jitter were stuck in the past. Throw enough bandwidth at the problem, was the theory, and it's just not a problem anymore. Literally everything that matters to anyone could be stuffed into packets, and if that failed to meet some QoS requirements, then the requirements were wrong. And nobody would care because it'd be so much cheaper.

Ultimately the latter approach seems to have panned out, but IMHO, more's the pity. I still miss the imperceptible latency of a good old POTS connection. The way we all step on each other in Zoom calls today is a special hell we could've avoided.

You can, and a lot of people do, have fully deterministic Ethernet connections, e.g. EtherCAT.

Oh I'm aware, but it's a field-bus, rarely seen more than a few hundred meters in diameter. You can't order an EtherCAT circuit with one end in Montreal and the other end in Dallas and maintain the same determinism. That used to be as simple as dialing a phone.

I very specifically used and heard used around me these exact verbatim terms and I can tell you the cultural divide was palpable between the two camps one from valley startups another from post monopoly AT&T. This influenced decisions on building wires all the way up to wan/transmission and standards. The Class 5 PSTN approach vs the best effort of VoIP Packet switching, IP as Ethernet everywhere vs ATM and Frame Relay nailed circuits, the ITU vs the IETF overlaps. H323 vs SIP is a perfect example of this tension and in fact the categories still exist. H323/Megaco/QSIG et al reference designs are precursors of the spaghetti of 5G/IMS etc and SIP is still the simplified out of band signaling template for many Web protocols.

Back then there was more friction because Internet and Data Comms was the new kid on the block basically disrupting a fairly established PSTN / Post & Telegraph infrastructure. And I say all this as NetHead who had many BellHead friends so it never got in the way of going out to lunch together.

This reminds me of the switch from circuits to packet-switched networks. Basically what happened is that telco operators only wanted to sell circuits which, like a circuit board, were essentially a closed loop. This is why t1’s were always dedicated 1.5mbps symmetrical links that were basically unshared bandwidth. You got 1.5mbps because that’s what att provisioned (until they started over provisioning the headend/noc and sold ‘shared T1’s’, but that’s a different topic).

What happened as I understand the lore is that one year a NANOG you had a bunch of young network operators basically get together and say “well, we’re done with this circuit stuff, we’re moving to packets where it’s a network of routers that make decisions about how each chunk of data is moved through the network based on headers.” The telco operators laughed at them because packet routing is much harder to guarantee deliverability of data than a circuit and they didn’t understand why people would want cheaper unreliable data routing for their internet service.

Of course, over time, packet routing has won for everyone except the largest orgs who have dedicated circuits that aren’t shared bandwidth. For anyone that’s used a direct 10GB fiber line that isn’t shared vs a cable modem that is overprovisioned to hell, the dedicated circuit is a whole different class of internet access (and 100-1000x the cost).

SIP vs dedicated trunks is kinda the same thing.

They did also sell X.25 (not circuit switched), but generally it was not competitive form a price/performance perspective with an IP network built from T-1s.

People wanted packet switched service, but found it was cheaper/better to build it themselves as an overlay network from circuits leased form the telcos.

> It was mostly an industry of carriers trying to charge as much as possible, while delivering as little as possible.

I remember when carriers upped the cost of text message from 15c to 25c per message. All four major carriers, iirc, changed their pricing within a month of one another. This is another service that costs them almost nothing. This is like one of those "trout in milk" evidence. I am positive they all got together and decided to raise rates but nobody went to prison for this collusion.

I mean I wouldn't have minded if they only charged for messages sent but they charged me for incoming messages as well. My friends had unlimited texting and as a poor college student I was trying to save money. I was so glad when grandcentral/google voice became available. I joined it and never looked back.

>I am positive they all got together and decided to raise rates but nobody went to prison for this collusion.

Do you think it could have been simple as they all realized that the market value of sending a text message had increased? If you recognize that a competitor is able to make more money by raising the price of something you also sell you can use that as a case study for why you should raise your own rates.

Theoretically it should be a market advantage to charge less (while still being profitable) than your competitor for the same service so you can get more customers.

It would matter how much of an advantage it would be. It isn't good enough to just be profitable, they want to find the most profitable thing to do. You would need to get enough people to text more / new customers to offset the potential profit you would have gotten from charging more.

At one point 2 weeks of SMS revenue alone generated more revenue than an entire year of all Hollywood’s.

What the OP fails to emphasize is that SIP was used to perpetuate the business model and all-important pricing. That’s why it was allowed to prevail: it did not challenge the status quo that really mattered.

In my country (Italy) now providers are starting to switch users from copper to fibre optic (and thus VOIP) at no extra cost. In a couple of years this switch will be mandatory, and the copper network will be completely dismantled (as far as I know they are already doing that in big cities).

For VDSL services (called also FTTC, fiber to the cabinet and then copper to the house) we already switched to VOIP, even if in theory both VDSL signal and the old analog signal, since it's cheaper to give to the users a router with VOIP integrated than do the conversion cabinet side.

Same in the UK - POTS (plain old telecom services) are being withdrawn at the end of 2025, copper will still be used for broadband but analogue phone service won't be supplied any more. Customers still using analogue phones will be given a device capable of converting it to VoIP locally.

The distinction was more of a research thing... it was the distinction between bell labs' networking efforts and the various internet groups, and in some sense very specifically between people like Sandy Fraser (who did the research behind ATM) and people like Jon Postel. The essential disagreement was where intelligence in the network lies... naturally teh bellheads (like Sandy) wanted a smart network and dumb endpoints, while the netheads (like Postel) wanted a dumb network and smart endpoints. I personally think the bellheads were right (putting intelligence in the endpoints is why the Linux Kernel needs such a massively complicated IP stack), but obviously that's not what happened in the real world, perhaps for the best.

Arguably, putting the TCP protocol in the Linux kernel is an example of bellhead thinking. The machines are now so large that in-kernel networking is no longer on the edge. Networking in the kernel on a machine with hundreds of CPUs—and hundreds of unrelated tenants—amounts to a smart network. Putting the network stacks in the user processes, and leaving the kernel as a dumb pipe, works better and is more aligned with the end-to-end principle.

I think bellhead thinking would be putting the network stack on (a possibly virtual!) network card.

Since you can never have "fully dumb" endpoints, this model requires upgraded end devices to make use of upgraded networks, and upgraded networks to enable newer end devices. That's an inherent economic disadvantage from the start.

Also, the computational and memory complexity of maintaining the state of every existing connection at least at some point in the network sounds baffling.

There absolutely were.

In the early 90s the telco guys were sure ATM was the true way, and IP was just gonna die off as it gained adoption.

I've heard Douglas Comer talk about how he used to get ribbed by the Bell Labs guys about "when are you going to come work on The Network, instead of that 'internetwork' toy?" There definitely was a cultural divide between the two groups, at least in the early days.

That said, I'd believe it if someone told me that that people kept talking about "bellheads vs netheads" divide long after the divide stopped actually existing.

Comer taught at my school and I interned at a company making hardware to locate in RBOCs back when local loop unbundling was a thing and there definitely was a divide.

The general thought of bellheads was that voip was much lower audio quality and that nobody wanted to have to reboot their phone.

Both are true, but people got used to lower quality with mobile phones, and it's not clear that they ever wanted to pay a premium for reliable voice quality, when they can get "probably good enough" for substantially cheaper.

I should also say that Comer had his own biases. He had an extremely dim view of ATM[1], but ATM is a great protocol if you want to keep POTS[2] latency low and still share the network with non POTS data. ATM was a good technical solution to the specific problem. Whether or not it was a problem worth solving is another question.

1: https://en.wikipedia.org/wiki/Asynchronous_Transfer_Mode

2: https://en.wikipedia.org/wiki/Plain_old_telephone_service

Oh, there totally were. I started my career at the CNET, France Télécom R&D. At one point they had proudly produced a glossy brochure extolling their “Web protocol without IP networking” project, essentially running HTTP over ATM switched virtual circuits, because TCP/IP had cooties or something. That was so many layers of wrong I can’t even begin to unpack it.

As late as the 70’s, you could get in trouble with the local telco for using an acoustic coupler on a residential phone line.

Reply on news.ycombinator.com