"Operating agencies shall negotiate commercial agreements to achieve a sustainable system of fair compensation for telecommunications services and, where appropriate, respecting the principle of sending party network pays."That's the input that a number of former monopoly, European telecoms sent to the member states of the United Nation's International Telecommunication Union (ITU). Why? Because, for the first time since 1998, the ITU will be negotiating new international telecom regulations from December 3 to 14 in Dubai. And that meeting will mark the first opportunity for the ITU to bring Internet issues under its umbrella.
Don't be fooled by the opaque wording above. If that request was put into practice, service providers get to prioritize certain types of Internet traffic and "the sending party"—a network generating content, like Netflix, Youtube, and maybe even Google or Ars—has to pay for the privilege of reaching consumers. Although opinions differ on whether free, unregulated markets are the universal solution to all problems, it's hard to argue against the success of this model in the case of the Internet. Even in places with limited broadband competition, the amount of bandwidth users get for their money has increased at rates far beyond those of any other industry.
A sine qua non for this success has been the fact that interconnection between different Internet networks is largely handled as a technical matter. It typically happens without money changing hands—and often without even signing a contract. This is the direct opposite of the phone world, where interconnection between networks is heavily regulated and the party sending the data does, in fact, pay.
Bringing interconnection between ISPs under the same regulatory regime as interconnection between phone networks would cause enormous harm. Even having such an interconnection pricing model for a new, better quality service would create strong incentives for ISPs to let their current service get worse than it is today. Think about how crappy service in economy class helps sell upgrades to business class in the airline world (theoretically of course).
Free riding leaves no capacity to invest
The European Telecommunications Operators' Association, or ETNO, is interested in creating a "sender party pays" system for Internet interconnection. This way, the big European ISPs organized in ETNO hope to open up a new source of revenue that will allow them to invest in their networks, avoid raising prices for consumers, and skip suffering in the market place.
In a CNET story, ETNO's Luigi Gambardella explains they want to keep the current Internet the way it is today, but they want to be able to provide higher quality service for a higher price. According to Gambardella, this higher price would be paid by the sender of the content.
Better quality of service makes all kinds of sense to people used to managing phone networks. Of course you need quality of service delivery; voice communication doesn't work if the digital audio samples don't arrive within a reasonable amount of time. You wouldn't want emergency calls to get a busy signal because telemarketers are tying up all the lines, right? (What about analog, you say? There's no such thing as end-to-end analog telephony. The phone network has been going digital for half a century now.)
Big ISPs have been grumbling about how the likes of Google and Netflix get a " free ride" for some time. ETNO's position is more subtle than that, but basically boils down to the same thing with some net neutrality issues thrown in for good measure. And it's not inconceivable that some developing countries—which have enjoyed a significant influx of foreign currency as a result of the "sending party pays" model in telephony—may want to extend this model to the Internet.
Let's figure out whether there are really any free rides to be had, and what the consequences of "sending party pays" would be. In order to do that, we need to understand how data interconnections work in the phone world as opposed to the Internet then dig down into some of the details of “peering.”
Interconnection in the phone world
Obviously, customers of network A want to be able to reach customers of network B, so all networks must interconnect in some way. In phone networks, when a call gets made from a customer of one phone company to a customer of another, the calling party pays. As the phone call traverses the network, so does (some of) the money. If Andy, a customer of AT&T's, calls Vicky, who is a Verizon customer, then Andy pays AT&T, and AT&T pays a little to Verizon for their efforts in connecting the call. This is "sending party network pays." It's a natural fit to the way phone networks work.
The "sending party pays" model allowed for the system in the US where there were local and long distance phone companies. Long distance telcos charge users for long distance calls, then pay a termination fee to the local telco of the person being called. In most places the importance of these termination fees has declined, but there are two main exceptions. In Europe, mobile networks charge steep termination fees. In some developing countries, areas receive a lot of incoming calls but don't have very many outgoing ones, so termination fees make for a nice bit of foreign currency.
Interconnection in the Internet world
Very little of this applies to the IP world. First of all, there are no "calls." Yes, most applications use TCP (Transmission Control Protocol) sessions, which are like phone calls in the sense that they have a start, and end, and information flows in between. TCP takes data from applications, puts it in packets, and transmits those individual packets across the network, retransmitting lost packets as needed. The majority of traffic across the Internet is TCP.
When a TCP session is idle, it doesn't actually use any capacity in the network. In fact, the big routers in the core of the Internet don't even know about TCP sessions; they just look at the individual IP packets. Also, users don't have that much control over the TCP sessions and IP packets their computers generate as the system and applications may communicate under their own authority. One thing is still the same: users of network A want to be able to talk to users of network B. In the case of small networks, this is simple: those pay wholesale "carriers" to transport their packets to and from remote places around the world. This is called transit service. Both content networks and ISPs pay for transit service.
The exchange of traffic between Internet networks of similar size typically happens with no money changing hands. This is called "peering." With Internet service, it's not just the party that initiates the TCP session or sends the packets that pays, but the party that receives them also pays. For instance, if you have a 2GB data cap, that can be 0.1GB up and 1.9GB down, 1GB up and 1GB down, or even 1.9GB up and 0.1GB down.
Back to peering. If Andy, customer of carrier A, sends 29Mbps to customer Bob of carrier B, then—assuming there's no other traffic—both Andy and Bob pay for 29Mbps. Because carriers A and B already get paid by their respective customers, there's no need to make life more complex by charging each other money for the interconnection. This nicely sidesteps the issue of whether A should pay B or B should pay A. This system keeps the bean counters out of the loop.
99.5 percent handshakes
The OECD (Organization for Economic Co-operation and Development) is an international organization "founded in 1961 to stimulate economic progress and world trade." It just released the report "Internet traffic exchange: market developments and policy challenges." In the report, the OECD marvels at the fact 99.5 percent of peering happens without a formal contract, based on nothing but a handshake. I can attest to this, having set up a peering session or two back in the day after nothing more than exchanging a few e-mails.
For small- and medium-sized networks, traffic exchanged over peering means money saved on transit service, so engineers are encouraged to set up peerings. For larger networks this is different, because those prefer to sell transit service to smaller networks rather than engage in peering for free.
Free rides: hot potatoes and early exit
However, there is a complication. Large networks typically interconnect in multiple locations. A router in network A in New York doesn't know if a packet ultimately needs to go to New York, Los Angeles, or Tokyo in network B. So it simply delivers the packet to network B at the earliest available opportunity. This is called "hot potato" or "early exit" routing. So when Andy in New York sends data to Bob in Los Angeles, Andy sends many large packets containing data to carrier A. Carrier A then immediately hands them over to carrier B, which has to carry the packets from NYC to LA, then to Bob. Bob periodically sends small acknowledgment packets, which carrier B hands over to carrier A in LA, then A then transports from LA to New York to Andy.
So it turns out early exit routing imposes the majority of the data transfer costs burden to the receiving network. Yes, it's true: when left to their own devices, routers would give content producers that peer with ISPs basically a free ride. The ISPs need many gigabits of long distance capacity for the actual data, while the content networks can suffice with only megabits worth of long distance capacity to receive incoming requests and acknowledgment packets. So the notion that content networks, whether they send out Web or social content, search results, or video (YouTube, Netflix), get a free ride is grounded in fact.
However, in the telephony world interconnection is often mandated by law or regulators. IP networks have no obligation to peer. It takes two to tango: only when both parties feel they benefit from the arrangement, peering happens. The big carriers very much want things to stay that way, so they've been very careful to have peering policies based on objective criteria, making sure to let sleeping regulators lie.
Many of these criteria basically boil down to "we only peer with people who are at least as big as we are" and incoming and outgoing traffic must be reasonably balanced. This means that pure content networks and pure access ISPs need not apply—those have 90 percent of their traffic in only one direction.
So, no free rides for content networks after all.
illing for interconnection
Although peering disputes happen, the system is both (cost-) efficient and effective. Imagine that we'd all be making 40 times as many phonecalls as in 1997, similar to the growth of the Internet the past 15 years. That would be one hefty bill every month. Having the telephony interconnection model imposed on the Internet would create big problems: carriers, content networks, and access ISPs would have to use massive resources to bill each other.
Last but not least, there is a fundamental limit to the number of phone minutes a user can generate, but nobody knows what high-bandwidth applications the future holds. If interconnection becomes a regulatory requirement in the Internet world, the balance of power will shift to the content networks (with early exit routing) or the ISPs (some form of sending party pays), or possibly even to the wholesale carriers depending on the exact details.
As such, the OECD report recommends establishing "a bright line" between the world of traditional telephony (where regulation is appropriate) and that of the Internet (where interconnection would happen on a voluntary, commercial basis). The report suggests regulators only step in when "a very high threshold of market failure" is surpassed.
Quality of Service
As mentioned above, it's completely reasonable to want a higher Quality of Service (QoS) for certain applications. However, giving time-sensitive traffic higher priority over other traffic opens a big can of net neutrality worms. The two main time-sensitive applications are coincidentally two that have brought in billions of dollars in revenue long before anyone heard of the Internet: phone calls and (live) video. In the past, each had a network specifically tailored to the application's needs—low latency for telephony and bandwidth guarantees for both.
The Internet, on the other hand, wasn't built around the needs of applications. Rather, it provides a lowest common denominator service that is easy and cheap to implement. While this works well for types of communication that can take network delays in stride, it's not an ideal fit for real-time communication.
Commodity or not
In important ways, IP transit is a perfect example of a commodity service, where the service you would get from different service providers is completely interchangeable. For organizations that have their own block of IP addresses, it's even possible to connect to two or more carriers and move traffic between them almost in real-time. However, there are three main objective measures of IP connectivity quality: loss, latency, and jitter. Latency is how long it takes for a packet to get from A to B, and jitter is how variable the latency is from one packet to the next.
Different applications have very different needs. Bulk file transfers want lots of bandwidth. Here, the number of lost packets has to be very low but latency or jitter aren't problematic. The Web utilizes many short transfers, which don't handle losing their first or last packets very well. With many short TCP sessions, the latency that the TCP connection setup packets incur also quickly adds up to user-irritating levels. VoIP works best when latency is low, but only if latency is uniformly low, so jitter is a problem. Loss, on the other hand, only leads to small dropouts in the audio and can be tolerated to some degree.
The OECD report argues that IP connectivity is not a fungible commodity in the sense that economists use the word. It is a "perishable commodity, like agricultural produce," because the longer the tubes are between two ends in a communication session, the more latency packets incur and the more bad things can happen to them along the way. It's similar to how fruit may get bruised and has a shorter remaining shelf life after being transported across the continent.
TCP performance takes a hit as latency goes up. Video streams are very vulnerable to being sent through over-saturated connections—it can take less than two lost packets in a thousand to make a high quality video stream unwatchable. So if Comcast forces Level3 to deliver Netflix's delicate produce to remote warehouses only reachable by narrow, bumpy roads, Comcast's users may be inclined to consume more of Comcast's own fruits and vegetables. (That is all if you wish to apply the OECD report's analogy to the peering dispute two years ago between Comcast and Level3.)
Outside the European network operators organized in ETNO and countries that enjoy a significant influx of foreign currency in terminating fees, there is wide agreement that free market conditions have served the Internet and its users well. Introducing regulations from the phone world would simply be detrimental. And some fear even more nefarious outcomes of the ITU's regulation efforts.
ETNO's call for additional fees paid by the sender in return for better QoS makes sense at first blush—but it creates awful incentives for ISPs. What if their current network is already very good? That way, the toll lane won't be in much demand. So it seems unavoidable that ISPs will let the regular service level languish in order to make the higher QoS service more attractive—to the detriment of every application that can't use the higher QoS service for financial, logistical, or technical reasons. If ISPs legitimately need money to invest in their networks, why not simply raise prices?
So dear ITU, please don't bill Internet use like phone calls. Not even the part that gets better QoS.
Dear ITU, please don’t bill Internet use like phone calls | Ars Technica