Hacker News Reader

We're quickly going to be in a world where all phone calls are VoIP and I don't think it's unreasonable to allow QoS tiers. I don't want to have my phone not work because my neighbors are torrenting. Hell, I already do this on my home network. Google just wants to make sure you don't have to pay to get in a QoS tier.

How about a corollary: why should your neighbor's torrents slow down just because you are making a bunch of phone calls? (And remember, people are using torrents for legitimate purposes, like how Blizzard uses it to distribute purchased copies of Starcraft II.)

The solution is to fix overall network capacity.

That's a 'fix' if network capacity were infinite. You can rephrase your corollary as 'why should my VOIP calls drop just because I'm downloading StarCraft II'. There isn't anything inherently evil in the notion that you might want to prioritize, say, latency-dependent traffic over traffic that isn't.

Haven't people done studies on this and found that the cost to implement QoS is greater than the cost to just increase network bandwidth? That may not hold forever, but it holds now in the U.S.

Current residential broadband lines in the U.S. could go much faster with zero technical effort. They're capped because Verizon and Comcast only give you enough bandwidth to make sure that you get more than you would with a competitor. They hold the rest in reserve so that they can raise speeds and please customers easily when a competitor raises speeds.

We've got the slowest Internet speeds in the developed world, and you can bet it isn't because of technological reasons:

http://www.websiteoptimization.com/bw/1004/

My point was that 'adding bandwidth' is not actually a solution to congestion-related problems. Double the bandwidth, now StartCraft II downloads in two instead of four hours, during which time your VOIP calls still drop. This is not unlike a problem urban planners have run across - adding lanes to a thoroughfare often does not relieve traffic congestion. So unless you can add bandwidth indefinitely so that there is never, ever any congestion, the issues are somewhat orthogonal.

Spoken like someone who doesn't understand QoS. QoS uses latency-aware networking to push low-latency packets closer to the top of the queue. That doesn't necessarily mean that they get full bandwidth priority. VoIP calls may delay Bittorrent packets by a couple tenths of a second. In other words, low-latency applications are usually inelastic, while file transfers are usually elastic.

Again: that sounds good, until you think it through.

What happens when an upstart wants to compete with Verizon on VoIP? Say they roll out an innovative new protocol that offers far better quality in a still-reasonable footprint. Or offers some sideband feature that existing VoIP doesn't. (say, streaming file transfer, letting you send data bits during a call while on the call directly to the person you're talking to. no need to open a mail client or sftp and negotiate a second connection)

Why wouldn't Verizon decide that this new competitor is a new flavor of VoIP that needs to be in its own QoS tier? It will have a different traffic profile, so you can't very well argue that they shouldn't be able to "optimize" it separately on technical grounds.

But now it's a new "type" and Verizon is free to kick it into the QoS round-file tier to either blackmail payment or simply hamstring adoption.

Maybe they only do that until they catch up and offer similar features. Or maybe they launch their own next-gen VoIP protocol in yet-another QoS tier, where the upstart is still not allowed.

Once you open the door to QoS by type --and leave type classification in the hands of the ISPs who directly profit from how these largely arbitrary lines are drawn-- there's a vanishingly small difference between QoS by type and by source.

> But now it's a new "type" and Verizon is free to kick it into the QoS round-file tier to either blackmail payment or simply hamstring adoption.

This is what Google's looking to make sure doesn't happen. No one should be able to pay for a QoS tier or kick a competitor into a lower tier.

It's not perfect, but as an end user I'd rather have SIP traffic prioritized and take a chance that a non-prioritized upstart standard will take longer to catch on than my calls to drop.

Obviously the best scenario is a dramatic increase in network capacity so QoS is a non-issue, but that's not realistic in the short term.

Your example doesn't make any sense. For purely technical reasons there would be no reason to send a VoIP call along with the "sideband feature" over the same connection.

Even if your VoIP application offers a feature like streaming file transfer, you wouldn't actually implement that by sending that data over the VoIP connection, because the VoIP data requires real-time performance (the call starts breaking up if packets are delayed) but your streaming file transfer does not. It would be silly to let your multi-gigabyte file transfer DoS your VoIP call.

If you simply opened a second, non-VoIP connection for the file transfer, everything would work fine.

Would Verizon have an incentive to cheat by favoring its own services? Maybe, but that seems to be what these Google talks are all about in the first place. I think Google is aware that any agreement would need more teeth than trusting the goodness of Verizon's heart to comply.

I think his point is that the examples are unforeseeable, not that his example is the one you should be worried about.

> "It would be silly to let your multi-gigabyte file transfer DoS your VoIP call."

1. I think it would be silly to assume a reasonable person would propose sending multi-gigabyte files down a VoIP connection. You might want to consider asking people if there was a miscommunication before assuming they're unreasonable or ignorant.

2. I thought the qualifier that it would remove the need to open a second app was sufficient to indicate that I was talking about opening a second app. Ergo "connection" meant the human-process of mapping a VoIP contact to a data-transfer app/credentials, opening second app, sending file, etc.

3. No, a second physical connection for the file stream doesn't necessarily mean "everything would work fine". Anything that requires a new protocol/extension would make classification of the product/feature/service game to be, well, gamed, by the ISP, for profit.

As the VoIP connection itself would need to at least be modified to signal/establish the physical data connection, you'd necessarily have a protocol variant - even if the file transfer itself was, say, ftp.

4. The whole point of the example is just "New product that extends old protocol, thereby giving ISP chance to QoS-round-file new product rather than compete". Focusing on the example itself is just wasting our time.

> "I think Google is aware that any agreement would need more teeth than trusting the goodness of Verizon's heart to comply." I think Google is doing nothing more than hedging against the possibility that legislation will undercut the FCC's net neutrality kick; particularly given legislative power being likely to tip back to the strongly pro-corporate side.

I sincerely doubt they're negotiating with Verizon over treatment of any traffic that does not come from Google.

> You might want to consider asking people if there was a miscommunication before assuming they're unreasonable or ignorant.

Your entire argument is based on the assumption that Google is being unreasonable or ignorant. You're posing completely trivial scenarios of Verizon gaming the system, and assuming that Google is negotiating an agreement that is vulnerable to such trivial gaming.

Furthermore, your only evidence whatsoever about these non-public negotiations is an extremely vague and misleading NYT article, and a paragraph blurb from Eric Schmidt. You know next to nothing about these negotiations (as do I), but you're completely confident in asserting that it "completely falls apart upon examination."

> I sincerely doubt they're negotiating with Verizon over treatment of any traffic that does not come from Google.

Did you even read what Eric said? "What we mean is if you have one data type like video, you don’t discriminate against one person’s video in favor of another." If what they're negotiating only applied to Google, it would be the opposite of the principle Eric is describing.

> "Your entire argument is based on the assumption that Google is being unreasonable or ignorant."

Absolutely not. It's based solely on Verizon and Google operating in their own financial best interests. In fact, as far as my argument goes, Verizon and Google are placeholders for any ISP and any content company. The argument would be the same if we were talking about Verizon/NBC, Sprint/Microsoft or AT&T/Time Warner.

I'm not arguing against some hypothetical secret terms of a specific rumored agreement. I'm arguing against the line of logic I quoted in my original reply. Who it came from is irrelevant. It's a very common justification for QoS-by-type and it's a wholly misleading simplification.

I only referred to Google's interests because you brought it up. It's not in their best interests for them to argue for terms from Verizon that benefit more than Google itself, because that would necessarily weaken any concessions they could get for themselves. It would be unreasonable to expect Google to go toe-to-toe with Verizon behind closed doors and sign a binding agreement for the benefit of the entire internet.

Ignore hypotheticals, look at history. Say you have a network and you decide that voice calls are so important you should have (real) 5 nines reliability, backup generators, redundant capacity, the works. Now let's say that people start using this voice network to send data. Let's say that people start using other networks for voice traffic. Let's say that people start sending voice traffic as data. Let's say that people start using other forms entirely where they would have used voice (even for extremely important business and personal uses).

How do you decide what gets priority? How can you even know?

This is what has happened in real life. People started using modems on their land-lines. People got rid of their land-lines all together and used cell phones or voip as their sole voice line. People started communicating and doing business via email, text message, even facebook. Million dollar deals have been hashed out over blackberry messages. Relationships have been made and ended over IRC and SMS.

Meanwhile, the land-line voice network is orders of magnitude more expensive to operate even as it has become less relevant, because we thought we knew how to prioritize communications.

Far be it from me to encourage the fostering of bureaucracy, but this is one case where a consortium with a self-governing body might actually be in order.

Imagine an industry-wide QOS board. The board solicits input and transparently sets standards for types of traffic and the priority they recommend giving them. Then providers who have signed on to the charter agree to prioritize exactly according to these standards. The board reconvenes annually to examine current technological trends and review and update their recommendations.

Possible? Palatable?

Sounds like J2EE.

Why don't we just call this QOS board the "FCC" and give it a congressional mandate?

While the problem your discussing is legitimate, your proposed solution is far from the only one, and I doubt it's the best one.

All of this will end up working out fine so long as money doesn't change hands in the process. It seems reasonable that someone will have to make a judgement call about whether a certain type of traffic deserves to preempt others or not. The really awful world will be if people can pay ISPs to influence that judgement call in their favor.

What seems to get lost in this debate is the difference between priority and type:

top) VoIP

2nd) Video Chat

3rd) IM

4th) Streaming media

5th) Web Browsing

6th) Large file transfers/torrents etc.

The idea would be to make it protocol "agnostic", and base it on the real-time requirements.

The most elegant, protocol-agnostic solution I've seen for doing this is 1st to throttle individual clients to their allocated BW (i.e. so that one client doesn't gobble up an entire shared link), and 2nd to simply kill any long-term (12hrs+) TCP/IP connections that are not over a recognized/allowed L7 protocol. The 2nd is very effective at starving unwanted bit torrent usage.

This method is in use right now by the guys who hang 5.8GHz antennas on grain silos, i.e. rural WISPs.

simply kill any long-term (12hrs+) TCP/IP connections that are not over a recognized/allowed L7 protocol. [This] is very effective at starving unwanted bit torrent usage.

I don't see how. BitTorrent will just reconnect to a different peer and continue.

I like first one. However, while the second one is fair enough for the example of rural WISPs - if my ISP starts killing connections as a matter of common practice then I'd be looking for a new ISP.

It seems to me that you're hitting on a likely issue that could let there be general agreement between Google and Verizon with great room left for negotiation.

In any case I hope they come to some sort of agreement, as imperfect as it may be. And shortly thereafter I'd like to see agreement that certain kinds of traffic, such as port scanning, should be filtered very aggressively. (Yeah, I can dream, can't I?)

Sure there are edge cases, but I fully support a decision that if it comes to dropping a call or interrupting someones youtube watching, the priority should go to the call. Video calls are an interesting case, but it seems to me that the audio stream should be given higher priority than the video.

But does that mean that a call takes priority over a live video feed? Does talking to your grandma take priority over a live feed of some major world changing event?

The key is to make the entire case moot. That can be done with a combination of software engineering and increasing bandwidth.

the challenge is how do you promote developers to build bandwidth efficient systems, as it seems bandwidth is now a distant thought in the minds of most developers.

Why should the priority go to the call? You don't know if the person is watching a Youtube video showing how to give someone CPR, and the person is busy dying. Would you prioritise a call from a telemarketer over that? Tele-medicine is just one example. Who decides?

I completely agree with roc that once you start thinking about it, there are just too many holes in the argument for prioritising on type. "Type" can just be a synonym for "provider". There will be nothing to prevent carriers to push new protocols to a non-priority "type", if they don't pay up.

If voip calls get priority, what prevents me from writing a torrent protocol over voip? That would just be a waste of bandwidth. Prioritising will set off a never-ending arms race.

As long as bandwidth is a scarce resource, something will happen. Either there's a well thought out procedure, or it's random. I think we can do better than random.

I argue that a "well thought out procedure" is not possible for such a complex system as the internet. Every kind of procedure that can be thought of will either have holes in the logic, or will be gamed by users and providers in a never-ending arms-race.

Emergent complexity is not "random", although it can be based on elements of randomness.

Nature is an example of a massively complex system that has evolved on elements of randomness, but the system itself is by no means random. And there's no way that it could have been designed with a "well thought out procedure".

I don't get your point. The pattern of dropped packets in a network will not "evolve" through some process of natural selection.

Can't you just hack a website page to tell it that it is a video page? I mean you can hack a php file to tell it it is a css file.

So if this can be done, wouldn't any attempt to prioritize data by time be futile?

Why should video be prioritized anyway? I hardly watch youtube, much less than search on google, read a website, do online research, or whatever else. I'd rather video was less prioritised. I can pause it for a few minutes while doing something else, but if a website has to load slowly if video is to be prioritized that has the necessary effect of lowering my quality of surfing, thus less google searching.