Hacker News Reader

In a lot of ways, modern voice calls-whether video or just voice-suck in comparison to old landline Ma Bell landline calls. It’s mostly a reasonable tradeoff but I had a hell of a lot fewer dropped calls in the old days.

You also had zero calls outside your house, in the car, or on the go :)

Hence "mostly a reasonable tradeoff." Even many of us who were adults pre-cell phones tend to forget just how tethered to landlines we often were. But we also mostly weren't yelling "Can you hear me NOW??" into our handsets.

> "Data could ride a voice-grade connection just..."

Perhaps this mindset was true at the AT&T / (eventual) RBOC level but at the R&D level I believe the story was different.

My father worked for Western Electric. For those who don't already know, WE's role was to take Bell Labs "raw" R&D and make it work commercially. In any case, when I was a pre-teen / teen (think early to mid 70's, if my memory serves me correctly) I remember going to an annual open house (for family members) an seeing demos on fiber optics.

Maybe the plan was to use fiber strictly for voice, but that feels odd to me.

p.s. Fwiw my father actually worked on a technology that then competed with fiber. As we know now, fiber won. My point is, there was a sense somewhere that additional capacity was necessary. That amount of investment for the growth of voice might be possible, but I don't think so.

Maybe the plan was to use fiber strictly for voice, but that feels odd to me.

If you've seen the eye-watering cost of 1600 pair copper bearers, not to mention the sheer time taken to joint it, you'd see why the telcos were keen on fibre for voice.

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

“SONET was originally designed to transport circuit mode communications from a variety of different sources, but they were primarily designed to support real-time, uncompressed, circuit-switched voice encoded in PCM format.”

Yes, I get the impression that the plan until ATM came along was to build exclusively circuit-based networks with most of the circuits used for voice and some for data.

Customers have gotten used to the latency of digital mobile phones, but some of us still remember the truly simultaneous conversation of a landline or analog cellphone

Right now in the UK there is a project to replace the emergency services radio network TETRA with a 4G based solution. Of course, it is horrifically late and over budget, so much so that it has well exceeded the costs it was supposed to save already. But its worst problem is the latency. On an old-fashioned system you can push to talk and say "don't shoot!" and be confident that's what the other callsign hears, but on the new system because of the latency all they will hear is "shoot!".

> On an old-fashioned system you can push to talk and say "don't shoot!" and be confident that's what the other callsign hears, but on the new system because of the latency all they will hear is "shoot!".

I know you're just using hyperbole make a point, but I very much doubt that the emergency services use voice commands as easily confused as "don't shoot" and "shoot".

According to this article https://www.theregister.co.uk/2015/01/08/airwave_tetra_switc... that is the example given by the Armed Forces as to why this is problematic

Push-to-talk latency isn’t a problem in the “it might replace SMS” scenario the mobile industry once envisaged for it, but it is in an emergency. The classic example the Armed Forces give is to imagine a commander who has a team of snipers on a roof pointing at a target. He gives the command: “Don’t shoot”. Unfortunately, in a cellular, IP based device, it takes a fraction of a second for the app to fire up and make a connection – a fraction of a second which is just long enough for the word “Don’t” to fail to make it into the message.

In my military training for the communications troops more than 35 years ago, we were drilled to "Think, Push, Breathe, Talk", because analog equipment has warmup latency, and besides, one would easily tend to unconsciously talk already while pushing the talk button.

If anything, it would be EASIER for digital equipment to work around this, analogous to how shutter buttons in digital cameras work: Run the microphone in always on mode and keep a short ring buffer of the audio prior to the button push.

That’s an odd example - surely the whole message will be delayed, and not cut off, unless you push the button before you start speaking?

Apparently delaying the message is rather hard for regular solutions. I've worked on a project where delaying the message (and playing a few tones first) was sufficiently hard that one of the reasons we got that project in the first place is because we actually could.

Thanks for the reference. I don't doubt that push-to-talk latency is a problem but I'm afraid my common sense still prevents me from believing that particular example, despite the Register's quotation of "the Armed Forces".

It may have delays, but there is no message cutting if you push the button before you talk.

Is there any way to get real-time voice connections over the modern internet?

Is it just a matter of setting up VOIP and then mercilessly debugging latency issues on both ends?

Is FaceTime or any other commercial solution close to zero latency?

Presumably a lot of the unavoidable latency (in any particular Internet path) compared to a dedicated switched circuit comes in when the routers examine packets, decode addresses, perform routing table lookups, execute routing algorithms, and then copy the packet into another interface.

If you had a physically switched circuit you could imagine that the data would travel over it at the speed of electricity in wires, while for packet switching there's some kind of software step and "hesitation" as the router "thinks" about the routing problem posed by each packet.

There are some Internet connections where the speed of light limitation dominates the routing issues, but I don't think this is the case most of the time when dealing with short-to-medium-distance connections involving multiple commercial ISPs. (Especially for peer-to-peer connections on residential broadband, you also don't have the latency benefit of a possible CDN node colocated with or direct CDN peering with your own ISP.)

So I think the best answer to most of your questions is "not really", although it would be really interesting to see a detailed breakdown in a traceroute of speed-of-physical-medium vs. various router tasks for each hop. (Then you could imagine improving some of these by changing the physical medium, reducing the number of hops via better or different peering arrangements, and getting fancier routers in some places that contribute less latency per hop.)

Periodically on HN we see pieces about the extreme lengths that people engaged in high-frequency trading will go to in order to get really low latency. Those folks know all the details of exactly where the latency comes in and exactly what could be done about it. But the huge expense and trouble that they have to go to in order to get some of these improvements suggests to me that we're probably not going to see the same improvements easily on the typical consumer broadband connection.

Decision time in the router usually isn’t an issue. That’s mostly a fixed cost that can easily be kept under a millisecond. Most of the time gets spent queueing for your turn to either come off of the interface or to go onto the egress interface.

All of your jitter in a stable configuration router is going to be coming from those varying queue depths.

Thanks, I don't really have any intuition for the relative contribution of these factors!

> people engaged in high-frequency trading

These people pay to have their servers colocated with the trading servers they are communicating with. That would be like having a VoIP call with someone sitting next to you connected to the same switch.

Though I doubt the difference between a 50ms vs 2ms latency would be perceivable for a phone communication.

If by "modern internet", you mean TCP/IP, then no. You'd need a circuit-switched network to do that, and those are all obscure/niche networks except for the Public Switched Telephone Network.

Real landlines work even if the power is out.

> as far as the phone company was concerned, voice was the only thing

I wish the phone companies would take another look at the way they handle voice calls. The entire experience is terrible. You can get better voice quality and less spam using just about any service (like Facetime or Skype) other than a standard voice line.

I don't know about that, Skype clogged up with random invitations and messages (when I last used it 6-7 years ago..) about as often as I get spam calls now.

I get at least two spam calls per day now. My record in one day is eight calls. I've started using Google's call screen feature and that makes it much more bearable, but really I wish I could have my phone give a busy signal to anybody not in my address book.

I get maybe four or five spam Skype invites per year.

> Radio studios still use ISDN links to other studios, because they have fixed latency and incredibly low jitter.

Can confirm. We have a good friend here in LA who is a voice actor (among other talents).

He still pays for an ISDN line at his residence to allow remote work — obviously, high audio quality is a must have for professional use.

At work I have our VoIP phones (and fax ATA) running over a cheap T1 connection. Never had a call quality problem. It's amazing.

I also remember the very high latency of long distance, international calls from the analog days.