Hacker News Reader

Comment by Wowfunhappy | Hacker News Reader

I really appreciate Mumble's dedication to lowering latency.

Latency is a silent killer of conversation quality. When it's relatively small, you don't notice it's there, but it makes conversations more stilted because everyone's responses are just a bit less immediate. You chalk it up to the person or situation, rather than the software.

I'm not seeing anyone make significant attempts to solve or reduce this problem. To my ears, 2010-era Skype had lower latency than any VOIP software in mainstream use today, including Skype. And of course, latency over POTS was fantastic, albeit with terrible audio quality.

View on HN

> To my ears, 2010-era Skype had lower latency than any of the modern VOIP systems in use today (including Skype).

Yeah! What happened?

Skype went from point to point systems to routing through a server backend.

Any idea why? Sounds stupid, considering latency losses and increased server costs.

So that the data can be captured, analysed, etc.

So the NSA could listen in.

Compliance

Well, it happened around the same time Microsoft bought the company. That's not an answer per se, but probably had something to do with it.

In fairness, Skype seems to be a lot more "stable" now than it once was. I recall having to frequently restart calls due to the audio suddenly crapping out in a variety of odd ways. That seems to happen much less often now. Can't say for sure if the two are related, but it would make sense.

But, I'd take a VOIP system with decreased latency over stability any day. Particularly since the problems could almost always be fixed by restarting the call.

Probably one of two reasons.

1. Control. They would control the medium, an thus have full control over the communication. Maybe even apply wiretapping if needed.

2. Anonymity. P2P usually means that you know the address of the other participants, with a central server you don't get to know their address. Professionals might need this extra layer of protection.

Hole punching through routers. It's hard to generically connect two NAT'd devices directly. It's simpler to route through a central server.

Hasn’t this been a solved problem for a while with STUN / TURN servers?

The centralized server just handles the negotiation, so that the p2p clients can do their thing.

I don't recall this ever being a problem with Skype, and I used it quite extensively at the time.

(Mumble is another matter...)

Skype's original USP was that it resolved the NAT issues that plagued SIP based VOIP implementations, exactly by performing hole punching through routers (I suspect just UDP hole punching)

This isn't that hard. It's even easier now that many people have IPv6. IPv6 still usually requires rendezvous due to stateful fireballing but it otherwise behaves like 1:1 NAT. (IPv6 is occasionally deployed with NAT, but in that case it is 1:1 NAT without port remapping which is also P2P-friendly.)

Surprised no one mentioned the documented reason: the VirnetX parent lawsuits, which were also why Apple had to stop using P2P in FaceTime.

https://arstechnica.com/gadgets/2013/04/after-200m-patent-se...

Apparently patents in this form need to be abolished ASAP. Human society can't afford anything like this standing in the way of widespread usage of such well known and well available technologies.

Please, let's all go and short that patent troll.

Apparently that troll also forced changes to Apple's handling of on-demand VPNs: https://arstechnica.com/gadgets/2013/04/apple-says-vpn-chang...

IMO, all patents need to be abolished ASAP. They're a bad idea to begin with, and this case is just particularly ridiculous.

Don't hate the player, hate the game.

Maybe for technology. The patents are still somewhat useful outside IMO. The whole idea about patent was that you could enjoy temporary monopoly in exchange of disclosing information about your invention.

For example thanks to patents we have generic version of medication, instead of them being trade secrets. Yes I know there are other things like drug approval that requires disclosing this information but that works together with patents.

Go back through history and you'll find otherwise.

Can you cite some knowledge and links for us, beyond "go back through history"? :)

Radio, light bulbs, steam, diesel, and Otto-cycle engines, television, software, pharmaceuticals, ...

And, more substantively:

https://www.sciencedirect.com/science/article/pii/S221256711...

Patents are fine when the invention is truly novel. However, we should be much more strict about what we are willing to grant patents for.

Software patents, yes. They never should have happened. Stallman was right etc. Other patents have very different behaviour and consequences. One of the worst things about "IP" law is that it causes people to bundle together vastly different concepts.

Can anyone explain to me why so many companies invest in VirnetX? it's pure evil. https://www.nasdaq.com/symbol/vhc/institutional-holdings

Money is the root of all evil, etc

"Love of money is the root of many evils."

It's a subtle but important distinction: money of any kind is a tool, existing for a valid purpose; otherwise inert.

Accumulating money in a manner viewing it as a score versus others easily leads to looking down on those who have less, and to practices that optimize return over anything else. This can dehumanize others in the eyes of a money lover.

https://en.m.wikipedia.org/wiki/Love_of_money

Most of the top holders of any big public company are funds that invest in every big public company.

Some big company should try to to buy up enough stock to control the patents.

> VirnetX filed suit in the East Texas judicial district that it favors

That sounds like good patent troll. Did they create any product using the patented technology or just Sue everyone?

In the online gaming space, Skype used to be a big enabler of DDoS attacks due to its P2P nature. Almost everyone used Skype back before we had Discord and other alternatives, and there were plenty of sites that you could paste in someone's name and have their IP in seconds. Even if they weren't online, Skype servers cached a bunch of previously seen IPs. Then they would paste that into a $5/mo "stress test" tool and the victim gets a short DDoS (1-5 minutes) but enough to disrupt / kick them out of the game.

this was done well after skype stopped using p2p routing

Primary reason not to use Skype back then. Ts/Ventrilo existed longer than Skype however. So a money issue then? But public servers were/are available too.

> Almost everyone used Skype back before we had Discord and other alternatives

Yeah except for when gaming, Skype "always on" voice was very annoying for any games and the latency was way higher than Mumble or Teamspeak so we would use those for our gaming groups but Skype for person-to-person calls rather than group chats.

I guess if you didn't have anyone technical enough to setup a Teamspeak or Mumble server in your gaming group YMMV but they were 1000% better than using Skype which had worse quality audio, no push to talk and huge latency.

See....Skype has had PTT...They've always had it...Discord is the biggest garbage app ever created It will NEVER be ANYWHERE close as good as mumble or hell even teamspeak

A lot of people are giving legal reasons but there were real technical ones as well in 2011 that forced the migration.

When skype was first launched 95% of users were on desktop or laptops that basically never got disconnected. Skype leveraged this by designating some users as "super nodes" that basically acted as the backhaul servers that eventually replaced them.

This worked well until mobile phones became prevalent and began to outnumber desktop and laptops on the network.

Mobile phones have two issues:

1) They aren't always on/connected so can't be used as "super nodes," especially because even today mobile phones suck at mulit-tasking, but were even worse back then.

2) Even if 1 wasn't an issue the short battery life of mobile phones meant that they wouldn't last very long.

At a certain point it became impossible to support the increased number of phones with the current infrastructure choices. So they moved to something else.

This alone doesn't explain why couldn't they leave the "super node" infrastructure in, with an extra bit of code ensuring mobile users are never designated as "super nodes". This way, quality would remain the same for the desktop/laptop users, instead of being degraded for everyone.

Not to mention that supernodes, though fewer in number, could still help reduce latency for those on mobile.

The switch from crowdsourced supernodes was mostly about link quality with:

- asymetric links (ADSL)

- WiFi

- usage-based plans

- people shutting Skype down on varous occassions,

- suspended laptops,

- bandwidth competing applications (video).

I remember past Skype had lot of quality issues, sometimes connection was very choppy, dropped unexpectedly, or took long time to establish.

I said this downthread, but I'd happily put up with those quality issues (and I remember them, they were real!) in return for the reduced latency.

Current Skype still has those same issues. I'm lucky if I go a day without some bizarre "I can hear you but you can't hear me" type bug or a dropped call.

Mobile phones.

So Microsoft ol buddy the NSA could gobble all that data from a central point.

For Skype specifically, I noticed latency get worse around the same time Microsoft switched from P2P to dedicated servers.

3 letter agencies demanding access to private conversations happened, so this is very likely the result of all traffic being routed in a more centralized and less efficient way.

If it can negotiate a peer to peer connection, it still does and it's the preferred route. It's only in the case that there are firewalls, certain types of NATs etc that the calls is hairpinned via a server.

P2P vs using central servers.

This is all thanks to NAT. Skype had very creative functionality to pierce through NAT.

One thing is just the significantly higher latency on consumer/business Internet connections. ISPs all optimised for bandwidth, not latency. They don't market latency. This is thanks to things like packet interleaving. Back in the early 2000s it wasn't uncommon to get a sub 10ms round trip to a server in the same country. Now you'll be lucky to get sub 30ms.

So we get things like HD video of the person's head bobbing around, but can't actually have an enjoyable conversation with them.

The distributed architecture was not conducive to spying on users.

Reply on news.ycombinator.com