Difference between revisions of "FTC NN2"

Latest revision as of 16:13, 19 February 2007

• See FTC Workshop Transcript for index

Technical Background Panel

This panel provides technical background to help inform subsequent discussions and cover topics such as: How data is routed over the Internet; the various segments of internet access, from the end user to the content provider; how data discrimination can occur, be detected, and be countered by users and content and applications providers; and congestion and potential capacity limitations on the internet.

>> CHARLES GOLDFARB
Welcome to the opening panel of the FTC broadband workshop. I'm very happy to be anticipating in a workshop for two reasons. First, I began my public policy career in the FTC's Bureau of Economics way back in 1974 to 1978 so this is a nice reunion for me. Second, in my current position where I cover telecom and media competition issues at the Congressional research service, it's my responsibility to help frame public policy issues and to provide balanced and nonpartisan policy analysis to my 535 clients, 435 members of the house and the 100 members of the Senate. Chairman Majoras, Maureen Ohlhausen and their FTC colleagues have correctly framed the fundamental issue as broadband connectivity competition. This is perhaps the most complex issue that I've faced in my 32 years in Washington, and therefore I'm particularly glad that the FTC wants to spend these two days developing a firm technical and factual base for the ongoing debate. We're lucky to have two panelists with us today, one by telephone I hope, Bill, are you there? Well, maybe we have one panelist with us.

>> WILLIAM LEHR
Yeah, I'm here.

>> CHARLES GOLDFARB
We'll try to figure out how to get a microphone to you. His flight was cancelled this morning. But we have two panelists who have a lot of experience bringing technical -- their technical expertise to bear on public policy issues. John Peha is Associate Director of the Center for Wireless and Broadband Networking, and Professor of Electrical engineering and Public Policy at Carnegie Mellon University. His primary research areas involve technology and policy issues of computer and telecommunications networks, electronic commerce, and technology policy. William Lehr who is up in Massachusetts right now is a Research Associate at the computer science and artificial intelligence lab at MIT. His current research with the communications futures program and previous research with the MIT research program on Internet and telecommunications convergence focus on emerge Pentagon broadband and wireless terminologies and their implications for industry structure, business and public policy. This opening panel has a relatively narrow mission. To provide a factual technical base that can be used in the various public policy discussions that will follow over the next two days. We face the challenge of information on the technologies available to operate and manage broadband access networks without bogging down the nontechnologists who are essential to the public policy debate. Our plan is as follows. First, John will discuss the technologies available today or soon to be available that allow broadband network access providers to discriminate or differentiate among applications or users. Then Bill will discuss the technologies available to independent applications providers and end users for counter strategies if they face discrimination or differential treatment. I will then pose several questions to the panelists and then we will take questions from the audience. So John.

>> JOHN PEHA
Okay. So I'm going to begin with a discussion of some of the underlying technology and then economic and policy implications. We have heard and will certainly hear over the next two days advocates saying, advocates of network neutrality saying that networks have the ability and the incentive to limit customer choices through discrimination today. We will also hear opponents of network neutrality say that network neutrality legislation could interfere with useful activities related to discrimination and unfortunately both of these are right. So I will talk about the emerging -- how emerging technology can discriminate. I will talk about why it is beneficial to users, I will talk about how it's harmful to users, at least if the network has sufficient market power, why we need to balance these things, and I don't think I have much time for this. Actually I can spend two days on this about how the issue has been misframed on big principles and away from where I think the issue is. So what is net neutrality? I don't know. I've been following this debate for a while and I don't have a clue. Definitions have not converged, the best I can tell. But if I go back to principles endorsed by the Federal Communications Commission a few years ago the consumers should have access to the legal content of their choice, be able to run applications of their choice and be permitted to attach devices of their choice, all three things related, I would argue, to discrimination. The fourth of receiving meaningful information on service plans, later choosing among competing providers is interesting, but I'll focus on the first three because they are related to discrimination. First of all we have to figure out what access means, and access to me can mean one or all of three things. Access could simply mean that something is available, it is possible for me to use the Voice Over IP application or it is possible for me to access that website I want. It could mean available at an acceptable quality of service or it could mean available at a reasonable price. So we'll talk about each of these three things, first to go back a little bit, the Internet is based on the concept of packet switching, that is we will take all information sent, we have to divide it up into little discrete pieces, each to each of those pieces we'll slap on some control information at the front and the back which we call a header and a trailer, kind of like you put control information on the outside of an envelope when you mail it. And each of these packets is sent separately. So based on traditional Internet technology, might have been a better phrase, delivery of these packets is entirely best effort. That is packets can be lost, packets can be delayed, packets can come but not in the order you sent them, and it's entirely up to the sender and the receiver to sort that out and to request retransmissions where needed. Traditionally most resources have been allocated on a first come first serve basis. Actually the protocol for 35 years has allowed priority but for the most part people haven't used it or even implemented it. In general there has been little intelligence within the network. The idea is to push the intelligence to the outside of the network and try to keep up with packets as fast as you can, which among other things means that there was traditionally little ability to prevent against security threats inside the network because it wasn't built in. This is changing. In many ways there is more intelligence going in the middle of the Internet and I won't talk about all of it except related to discrimination, and discrimination I would argue has two components. First, you have to decide which packets or users or streams you want to favor or harm and then you have to figure out what it is you want to do to benefit or harm them. So beginning with the first, how do you determine which streams to favor. The traditional way to do that was to look at each of these packets as they went by, look at the fields and the header, that control information that was tacked on, one packet at a time, and things like IP address and port number, you can learn who the sender is, you can learn who the recipient is. Sometimes you can learn who the device manufacturer is for that device at the edge depending on where you are monitoring. Once upon a time you could learn who the application was through something called a port number but that hasn't been reliable or meaningful for a number of years. But some new methods have not emerged, actually they are not that new. They've been in universities for a while but they are actually products on the market that do new and interesting things to differentiate among packet streams. One is called flow classification. Something new, it will actually keep track, it is called stateful, I will keep track of every stream that is going by, my monitoring device. For each of these I will keep track of things like packet size, the time between packets and stream duration and I can learn a lot about the application that way. Even if you encrypt it I can. I can also do something called deep packet inspection where not only will I maintain state for every stream of packets going by me, I'll actually capture some of those packets and I'll reassemble them as if I were an application. I'll take a bunch of your packets together and I'll reassemble that email message if I want to. And in fact, if I'm doing that I can actually go a step further. As long as I've got state on every session and I'm pulling this information, I can also use this to cross index with other information I might have, like your billing information or your credit information or whatever you want. When you put all this together, I can have a really detailed information about who you are and what you are doing. I know the subscriber, I know the application, I know the content and the content or service provider. I often know who made the attached device and billing information and all the rest. So what might I do with all of this detailed information? There are a number of ways I might use this to discriminate. One is I may simply block streams, drop all those packets. Another, I might do is I might divide the traffic into channels and channels can mean a lot of different things but I'll group them together here. And some of them are better than others. What is particularly interesting about this from a policy perspective is I'm not sure this meets the traditional definition of discrimination but it certainly has an effect of giving some better service than others. A third thing I can do is I can use a wide variety of traffic control algorithms to adjust data rates, to adjust end to end delays, to adjust packet loss rates or blocking rates, that is entire streams that are not allowed to start. For example, the scheduling algorithm which says there are a bunch of packets waiting to go. Who gets to go next, or a dropping algorithm which says the buffer is going to overthrow, what do you throw away? Or routing algorithms, which paths should this packet take. And I can if I want introduce discrimination into any or all of those. A fourth thing I can do is sometimes called content billing or content charging. I can look into your packets and I can decide to adjust your bill. Up or down based on your application, based on content, based on subscriber. This is actually in fact a little easier than managing your quality of service in realtime, is to adjust your bill. So what do I do with all this stuff? Well, first let me make the argument that discrimination is wonderful, that I can do really useful stuff with this. One thing I can do, for example, is I can watch for security threats and block them. I can watch for viruses. I can watch for denial of service attacks. I should caution. Some of the proposals actually had a card out for security. The hard part is defining what it is to exempt security, what that means, and particularly where I mean I have students right now back at Carnegie Mellon University who are using deep packet inspection to find security threats, particularly spyware, we're trying to develop some new techniques. I can also tell you there are false positives and there are false negatives. If I block something that I'm 95% sure of security threat, am I going to get find 5% of the time. It's tricky. I can also block traffic from nonconforming devices. It's a way to make sure that all of the devices who are here actually obey the protocols and not cause problems for their neighbors. I may want to discriminate to improve fairness, particularly with always on connection traffic for a very small number of users can dominate the network and starve everybody else out. Peer to peer is a problem. Some may say after you reach your monthly limit perhaps I should block your traffic or give it a low priority or just charge you extra for consuming all these resources and that will prevent starvation of others. Another reason why you might want discrimination is to support diverse services. You'll sometimes hear that a bit is a bit. It simply isn't true. Not all bits are created equal from a network engineer's perspective. Some put more of a burden on the network than others. For example, if one application produces a steady stream of bits and the other produces big bursts, the burden per bit is different, or if they have different quality of service requirements the burden per bit is different or depending how they adapt to congestion the burden is different. If my traffic control algorithm is descriminated it turns out I can carry more traffic and still meet quality of service requirements which might reduce infrastructure costs per user. And if I discriminate in pricing, one thing I can do is get you to accept a lower, a lesser quality of service if you don't mind, no one is going to say I'm willing to tolerate low delay, give the other guy priority unless you give him a price incentive. You can also give incentive to shift usage to less congested periods and in general you can align price per packet with cost per packet, just sort of a complicated concept here but cost in general here is opportunity cost of not carrying something else. So that's why discrimination is wonderful. Why is it terrible? A couple of reasons. First of all, as you might expect, if, assume for the moment I have a monopoly in some part of the Internet, presumably the last mile connection. I now have extensive information on who you are and what you are doing, and I can use that information to try and set the price as close as possible to what -- to how much you value the service, to your willingness to pay. And economists will tell you that you are then extracting the consumer surplus, you are shifting benefit from the users to carrier. And users in this case means both consumers and content or service providers. I can do that pretty effectively -- the more information I have the more effectively I can do that. I will also probably intentionally degrade quality of service so that those who value a better service will pay for that better service. In fact, we see that coming. Nothing surprising here. This isn't the broadband market, the transport of bits. If you have a monopoly in the broadband market I expect you are going to try to get monopoly rents in the broadband market. What is really interesting is you may move into other markets. There are many upstream markets or some people say downstream markets. But in any case, markets that depend on the Internet for their existence, electronic commerce, communications like video conferences or Voice over IP, information distribution, whether it's video streaming or MP3s or something else or online advertising. And I can try and affect those markets as well. Just as with the broadband market, I can exploit the extensive information I have, I can deliberately degrade quality of service to further segment the market where that's useful, and I can try and set price as close as possible to what consumers are willing to pay in each market now. So I may want to separate, it's not -- I won't treat all packets the same, I will separate the Voice over IP market with the download of digital products. And within the latter I will separate a 4 megabyte PDF from a 4 megabyte MP3 music file. I may differentiate one song from another. So, for example, without, if I'm a carrier, without offering any E- commerce services I can essentially tax E-commerce. I can say I'm going to tax, put a 1 cent charge on book sales and a 2 cent charge on CDs. Why more for CDs? They are exactly the same but I'll charge what the market can bear. I may put a tax on ITunes and maybe even differential based on the popularity of that particular song. And if it's particularly interesting I might put a tax on Voice over IP, 10 cents per minute on Voice over IP. What's particularly interesting about that is that if you -- you can turn what is possibly a low price alternative to a high price alternative which is useful if you are offering telephone service. So some observations. I can use these techniques to protect my legacy services, that is telephony for DSL provider or video for cable provider. I can try and extract monopoly rents from competitive markets, that is the consumer may pay the equivalent of monopoly price even in the competitive market, which I'm not sure that meets the definition of anti-competitive practices, which is where our antitrust comes into play, so I'm not sure how this interacts with current law. And I can do this without entering the market or affiliating with a provider. So people who talk about this as an issue of favoring affiliated versus nonaffiliated content or service providers, there is more to it than that. There could also be content filtering for other reasons, perhaps for political reasons I will want to limit access to advocacy groups for issues I oppose or candidates I oppose, for commercial reasons I might want to limit access to commercial rivals or consumer complaints or labor unions. There is a slide missing from this slides which says there are accusations that some of this is happening already and there are denials. I don't claim whether it's happening or not but it's certainly technically possible. So where does that leave us? It leaves me in the hope that we can find a policy that, what I call a balance policy, that prevents networks from fully exploiting market power, from using discrimination in a manner that limits discrimination to prevent them from fully exploiting market power to seriously harm users, but does not prevent them from using discrimination to greatly benefit users, which may not be simple. I would conjecture that the impact on upstream markets is probably most important for the serious harm, and one observation is that when you are doing that you might see prices inconsistent with costs. For example, the price for carrying Voice over IP might differ greatly from the cost preparing for carrying Voice over IP and that might help us figure out when the problems are occurring, but it remains to be seen. So some conclusions on discrimination. Discrimination can benefit users greatly, improve security, improve quality, decrease infrastructure costs and allocate resources to those who value them the most, and so imposing network neutrality, a policy that prevents these things could do real harm. On the other hand, discrimination can harm users if the network operator has sufficient market power. It's because the network has access to a great deal of information, and it can use this information to discriminate, to extract consumer surplus in both broadband and upstream markets, even if the upstream market is competitive and even if the network is not affiliated with any upstream provider, so not imposing network neutrality could do real harm. My final conclusions on network neutrality, I think that means we need to focus on the specifics of a balanced policy. I don't hear a lot of talk about specifics. You know, can we deter the most harmful and allow the most beneficial? I don't think it will be necessarily possible to eliminate all harmful and preserve all beneficial and therefore really strict litmus tests like that are probably going to get us into trouble, too. It's a little more subtle than that. I can argue and I don't have time, the debate has been misframed, it's not about the inherent of discrimination, it can be useful, it's not about unfair relationships. It can be broader than that. The right of networks to differentiate or the freedoms of end users are interesting concepts but they don't provide us enough guidance as to what really should be allowed or not allowed. And for those who want more, there is a paper with much of the content I've just presented. Thank you.

>> CHARLES GOLDFARB
We shall now experiment to see if we can hear from Bill. Hopefully this will work and if not, John will be helping out. But, Bill, are you ready to go?

>> WILLIAM LEHR
Yeah, I am. I'm here. I'm in Concord and my great apologies for not being able to be down there virtually. Can you hear me?

>> CHARLES GOLDFARB
Can the audience hear. Yes, it seems to be working. I guess what you may have to do is inform whoever is moving the slides when you want to move to the next slide.

>> WILLIAM LEHR
Why don't you get my very first slide there. Let me again apologize for not being there. They cancelled my flight in anticipation, I believe, of the weather. It doesn't seem particularly evident, certainly not up here and from what I understand not down there either, at least that sounds good for you folks down there today. I am actually watching this on the Web at the same time to look at my presentation slides on the screen, so this is a strange occurrence but demonstrates, I think, the critical importance of the Internet, and its value in this situation. Let me say that by first by introduction, that I am an economist by training who lives in an engineering school which means I'm constantly confronting my ignorance on both sides of the issue, and that the paper that's going to inspire I talk today, you will notice is a joint work with Sharon Gillette, a former colleague of mine at MIT. Marvin Sirvue (ph) and John Peha also of Carnegie Mellon. So luckily John, the real engineer in this panel today will be there in live person, and hopefully will be able to take over and answer questions that I can't. Next slide, please. What I want to do today was first talk about my vision of the future. By that meaning, economists look at where the technology trends are taking us, and what this means for the future environment that we're going to be living in. To talk a little bit about why it's reasonable to believe there might be a problem about net neutrality and addressing that problem. I then talk about the joint paper that I wrote with John Peha and my other colleagues about what we call the net neutrality arms race and the sorts of responses that end users can have, and then ramp up with a brief, just, discussion of where I think that means about the policy agenda and about further technical work that needs to get done. Next slide. And if you can click through the whole, to the end of the slide, because I've put animation in these slides that is particularly difficult in this situation. Okay. So a vision of the broadband future. I think that it's pretty obvious to anybody that's engaged in work on the Internet following these technologies, following these industries, that the future of the whole information communications technology value chain is heavily dependent on the Internet and the future of the Internet is a broadband wireless internet. If you look at the really big things that have happened in this space that have been the sort of paradigm shifting changes that have driven major growth, it was the growth of the Internet in the 1990s, mobile communications also in the 1990s, delivering on the firsthand mass market data communications services, and then on the second hand, mass market personalized communication services. And then starting in 2000 the growth of mass market broadband, which really sort of unlocks the capability of the Internet. In the future in computer science speak, we're moving towards a world of pervasive computing. We already have computers in our cars, in our consumer appliances, in all kinds of things that we're not even aware of that are always on, and all of this computing power is much more valuable and useful if it's everywhere connected and increasingly a lot of that connectivity is going to be unaware. That's what we mean by the world of pervasive computing. We can also see this coming in things like RFID and sensors, smart network edges and the emergence of post PC device, all kinds of things that have computer chips in them that are communicating on our behalf, that businesses are using and increasingly consumers will be using that we may not even be aware of that are taking advantage of all this Internet. That means we're going to have lots of networks, and that no one size all solution or treating or thinking about these networks is either desirable or possible. You are going to have wired networks of many different types, coaxial cable, copper and fiber and you are going to have lots of different sorts of wireless networks from the WiFi networks that there has been a lot of talk about, to the 3G and 4th generation mobile types of networks, WiMAX technology, ultrawide band, free space optics, different kinds of satellite technologies, on and on and on. This heterogeneous technology is a characteristic of the future environment that's being driven by convergence and the need for interoperability and connectivity, but it will also pose challenges for all of those things. You are going to see a much more complex competitive landscape where the definition of who is a carrier, what constitutes a carrier, what service markets they operate in, making those definitions in a clean way is going to be increasingly difficult. And that broadband is really local and more local than traditional Internet access has been, because you are going to have very different sorts of environments that are going to acquire and make possible different kinds of technologies, certain kinds of wireless will work in places like the west but won't work in the heavily treed and more rainy northeast. Also, markets are going to differ significantly in their ability to attract and sustain infrastructure competition. Some markets are going to be lucky enough to have multiple fibers passing every home in the market. Other markets will be lucky if all they can get is some sort of wireless technology. And you are going to have overlapping generations of technology because the pace of change in this sector, if anything, has accelerated so that the differences of the -- and the issues that are going to be relevant in terms of what broadband looks like in one market may be really different, even across town in the same market, because of the, you know, terrain issues, what, you know, legacy infrastructure was available, etc. and finally we need a lot more investment in last mile access networks of all sorts all over. Okay? So I think that's the sort of technical future that when you think about policy in the net neutrality debate, you really have to be thinking about when you address this. Okay. Let's go next slide, please. So is it plausible to believe there is a problem? Well, first off, as we begin, as we've sort of gone over the cusp and increased the capacity of broadband connections, broadband traffic is growing exponentially. Before we had significant amounts of broadband access, you know, sort of pre-2000 the fact that most people who access the Internet were still doing it over dial-up connections throttled the ability of individual users' demand to reveal itself as how bursting and peaky indeed it can be. The services John talked about, pier to pier, different kinds of rich media, gaming, interactive media means that the Internet is having to handle a much wider array of traffic types, and a much greater volume of all traffic types that have different tolerances for the quality of service needs. So, for example, you know, Voice telephony is very sensitive to delays and so if the packets don't get through in a particular period of time the service is effectively unusable. Other services like email are much more robust, obviously, to delays, although even email is subject to congestion. And there are questions about how traffic patterns are shifting. Is it a few heavy users that are basically consuming way more resources than they are effectively paying for, or is everyone sometimes a heavy user that needs to burst because of the nature of the applications? There is not a lot of great data in the public sphere to make informed policy decisions about that, and even the carriers don't really know what this broadband traffic is going to look like because this is a growing and emerging phenomenon. Users are learning how to use these broadband networks, and as they use them they change their behavior. And as their behavior changes the carriers are finding they are having to address ever new challenges for managing this traffic. Another important issue is that penetration saturates and so revenues grow flows. So the question is if we want the industry to continue to meet the growth in traffic we have to figure about what the incentives are. There are a number of kinds of solutions that we may look at and all of these have problems with them. You can look at different kinds of traffic quotas and those are potentially an issue -- let me just talk to the next slide to catch up where you guys are. Okay. So anyway. Hopefully what I've tried to explain thus far and we have a paper that we've -- that talks about this in a greater way, is that indeed there is a real problem for the continued exponential growth of the traffic and the markets' current attraction to sort of flat rate pricing that means that provisioning continual investment to address the real growth in traffic is not an obvious outcome that's going to indeed happen, and that if that were to be forestalled I believe that would have deleterious effects for the whole Internet value chain. So let's go to the next slide. So scenarios for network nuetrality Arms Race, the reason we call it this is because we believe that there isn't an obvious outcome, that whatever efforts occur, let's assume has the power to discriminate might undertake, there are responses that the end users could do to that that in turn would induce further responses from the carriers and so on. So in this paper what we looked at, we said let's assume there is no net neutrality regulation, i.e., let's ignore any kind of regulatory policy interventions that might discourage the sorts of behavior that John suggests might be possible by a carrier with sufficient power and capabilities. And that indeed those things are done to discriminate. I put "Discrimination" in quotes to move away from the loaded term of, you know, as an economist may think about it or as folks may think about it as a pejorative bad thing that may be done. Just to say, just to focus on suppose a carrier does something a net user doesn't like? What can an end user do? Okay? So when we think about this we say why is net neutrality, why is net neutrality a concern? Can we go to the next slide? The fear is that they are going to engage in this -- click through the slide please. This has a lot of animation on it. Basically a lot of the points here I think we've already addressed. The fear that motivates the concern for net neutrality is that these carriers will block access to content, will offer differential quality of service, or will price discriminate. John has explained how that can happen. He's also explained that the ability to block access to content may be useful not just detecting and protecting against distributed denial of service attacks or viruses or other sorts of malsoftware, that differential quality of service may be useful and required for traffic management and that price discrimination may be useful for recovering of some shared costs of the network, what economists call Ramsey pricing, something you'll hear more about later in the day. So the question is is what they are doing really discrimination? You know, if there is really no problem, I don't feel most people have a problem with a carrier trying to recover the higher costs for more resources used, so, for example, if you are getting preferential caching for your service, your video service and that's costint the carrier more the carrier ought to charge you for that. If you don't care about that because you are providing free content, then maybe you shouldn't have to pay for that preferential caching. Similarly there's the literature on two-sided markets which I'll leave to folks later today to talk about. So the goal needs to be to protect against harmful discrimination but there is lots of types of traffic management that are not likely to be harmful, and it's important to note that in the sorts of responses we talk about here today, this is sort of one of the insights we gained from writing this paper, that end users' ability to respond doesn't really matter if what carrier is doing is actually something socially we like or not. End users who can have ability to respond may respond against anything they don't like. So would just paying higher prices or tolerating reduced quality of service when they can do that by sort of hiding, you know, their capabilities and that sort of stuff. So there may be another kind of concern here that hasn't really been talked much about which is the concern of what do you do about end users who are sort of doing the end run around good management practices in the Internet. So what are the kinds of responses that an end user can have if a carrier does something that an end user doesn't like? Next slide. There are three sorts of strategies here. Click through all three of the points here that I'm going to talk about quickly. The first is strategy one is they can do something to try and bypass the actual differentiation. In other words, the carriers attempt to charge higher prices or offer lower quality of service, the second sort of strategy we talk about are end user counter measures which are sort of actually trying to deal with the in band discrimination techniques using end user based strategies. In the third one we call learning to live with differentiation which is basically where it's just sort of using other aspects of the full Internet connectivity PI to effectively mute the impact of any discrimination by carrier and thus render it nonharmful. Let's go to the first strategy, strategy one, bypassing differentiation. Next slide. The most obvious way that you can get around the problem is take advantage of multiple bit paths. Now, if there are facilities based competition, that may be sufficient to render the whole concern over nonneutral treatment by a carrier moot. So that, you know, as an economist I would believe that if there is adequate competition the competition would result in carriers offering consumers what they want, and so a carrier that tried to abuse consumers, do something they didn't want, would find those consumers switching to other carriers. But even in the situation of where there appears to be ample competition in the originating market as John explained and I'm sure other folks will talk about today, there may still be a terminating problem where an individual end user doesn't necessarily know what content providers or application providers upstream had to go through to get to that end user consumer. And because the end user consumer doesn't directly pay the cost of that, he may not really care and may not be willing to vote with his feet to move to another carrier if that carrier is engaged in such activities. One-way to do this is if carrier -- if the end user is able to multi hone. For example, if it's a business, the business may actually, and this is typical of a lot of businesses, have service provided by multiple carriers, and so the content provider can go to that user across any of those because the user has those. That's less likely to be an option for the typical mass market customer, although in the future that may become a little bit more of an option through things like cooperative access sharing and things like scalable mesh networking. And there are ways in which, and you know, we know of situations of folks doing this already today, although it is at this point technical that is able to do this, but where people are doing things like I have a Comcast connection and you have a DSL connection and the two of us are able to share that because we're on the same local area network that we've set up, and so we now have routing diversity to get out to the Internet, and we have a way to actually share that. You can do even more interesting things like that. So upstream aggregation and consumer networks are a way to do this. A second way is broadband resale, so different types of technologies and uses that allow broadband connections to be shared more generally can help here, and there are different sorts of models that a number of folks have put forward for how this can happen, and we're sort of seeing experiments with this in the marketplace. And then finally by end users sort of, you know, in a much more concentrated way organizing alternative access connections, and municipal networking where communities get together maybe with the help of their local government, but usually with the help of their local governments or local utility, but not necessarily get together, provision a network. And if that network is an open access network, then that provides another way to deal with this. And we strike the importance of it being considered as an open access network because otherwise it's just another network, and so in principle that will help because more choices is better, but it's possible that the municipal network, if it's not an open access network could also be guilty of nonneutral treatment. There is no reason to presume that your municipal carrier, if it has market power, may be any better behaved than an investor run carrier. Next slide. So the second class of strategies are end user counter measures, and we sort of organize those into nontechnical and technical. The nontechnical strategies are if a bunch of end users don't like something that a particular carrier is doing, in the Internet space they've demonstrated a remarkable ability to organize and bring serious consumer pressure on this. We call this shining a light on the rats. So if there is a particular behavior that a carrier is doing, some sort of quality of service differentiation, it really has no justification in cost and looks really high handed, it's very common for this to get, you know, blogged in realtime and for this to embarrass carrier so the carriers, the operators, and force them to change their behavior. Now, is this something we want to rely on absolutely? No, but I don't think -- perhaps not, but I don't think we should neglect it when we think about the power that's all underestimated. Another sort of response is the ability to sort of rely on applications. A lot of the discriminatory techniques, I'm not using discrimination in a nonjudgmental way here, are attempts by the carriers to get users to self- classify so they say if you are a business user, then tell us and you'll pay more, and the reason you pay more is because as a business user we expect you to use more expensive resources. But if you don't really want to pay more you don't -- you just sign up for a residential DSL line and run your home business on it. I think most small home businesses that's exactly what they do. They don't opt for commercial services, and maybe whether or not they should or shouldn't, you know, is another question. But the ability to sort of misrepresent your user behavior in very nontechnical simple, you know, sort of old world ways is another sort of end user countermeasure. And of course, you know, if there is some of these sorts of behaviors are more likely to be able to work if the end users, if the discrimination is wide spread as opposed to, you know, idiosyncratic or distributed. There is also a lot of sort of different technical options. The paper goes into these. The technical options really depend on the level at which the blocking is taking place. So, in other words, is it happening at the application port? In other words, in some of the blocking, for example, in peer to peer applications is based upon identifying the port's use by those applications, it's relatively easy and the people doing applications in this space have done this where they can use the -- they change the ports to use ports used by common applications, that nobody really wants to block or by doing things like port hopping where the applications changing randomly the ports it's using which are attempts to offset things like application port blocking, and these sorts of quick fixes to programs can be downloaded and virally spread across these peer to peer programs very rapidly so it's not much of a burden, you know, to end users in a day of automatic software updates to keep abreast of these kinds of responses and sort of continue playing in the game. You can also do things like source and destination address filtering, you know, and traffic analysis based filtering to change the nature of the traffic you are offering to the Internet, if you are doing this upstream or by going through some sort of thing that objection secures this information. So there is all kinds of things like that you can do. One of the things that needs to be focused on is whether or not the discrimination that is being offered by carrier's quality of service, enhancing or degrading. For example, if what they are giving you is a higher quality of service for a higher price, it's very hard to get that by hiding your, the nature of what you are doing, unless you pay more. The other end of what they are doing is they are degrading your traffic if they can figure out what it is, then these responses are more effective. So the quality of service enhancing types of discrimination are much much harder to respond to by these sorts of technical end user counter measures, all of which we talk about essentially rely on hiding the basis of the discrimination. Next slide, please? The last category of technical responses they'll talk about are learning to live with the differentiation. By this we mean effectively suppose they discriminate and nobody really cares. It turns out there are a lot of applications that are just not very vulnerable so delay tolerant applications are applications with lots of substitutes don't seem to be particularly good candidates for concern about discrimination, so, you know, in a number of cases the postal system offers a good alternative to broadband delivery. We see the example of that in the case of Netflix versus online movie delivery. Netflix has crafted a pretty good business by shipping around CDs and a number of computer scientists have fondly quoted as reminding people that one should never underestimate the bandwidth of a bunch of tractor trailers loaded with DVD ROMs. There are a lot of kinds of options you can do, and a lot of applications and business models for delivering services that need to be considered when one thinks about, you know, how much these would actually harm individual consumers. One of the kinds of strategies, broadly one of the kinds of strategies someone can do here is buffering. In other words, they stream its technology at a slower rate than was really, let's say required by the broadband and they store it on digital video recorder, and then their ability to view it at whatever quality of service or capacity or rate that they want isn't affected by the service they are getting delivered from the networks. So they are using whatever the plain vanilla level service is and then they are getting the high quality experience. This will work for any applications that aren't real realtime. For example, that works for a lot of television, in a lot of -- I mean a lot of video entertainment viewing experiences, but not for all. It won't work for real sports for most people. It will work even for some people but probably not for that. Certain other kinds of programming like old reruns or something, you know, those are things that may be really not dependent on having very realtime access. But of course, the ability to do strategies like this isn't going to work if carrier controls your set top box or your digital video recorder, in which case that's just an extension of the network. And another thing that can happen here is if you are preloading contingent content, in other words, this is content that you might want to watch but you are not really sure, then this sort of end user response puts additional stresses on the network because you are loading traffic that in effect you don't really need to load and you are doing this because you don't want to deal with the fact that the quality of service you'll experience may not be what you want it to be. You can also do a lot more with distributing caching. In other words, capture traffic and keep it local. If someone in your neighborhood was viewing a movie, cache that locally and other people have it available locally to view. You know, the question is for what types of content will this work? It won't work for really live content but obviously it makes sense where it works. And so I think you'll see a lot more of this. And then finally there is different kinds of end user processing substitutes for conduits. So broadly computing communications and storage are all substitutes for each other along some dimension. And so if you have more limited transmission capacity you can use more processing to compress those streams and get by and get an equivalent experience. Again, this costs money by having fancier boxes at either end of the connections and you may have some degradation of quality depending on exactly what you are doing, but those are the sorts of things you do. Next slide. Click through this. So what have we learned? What we learned from this exercise is that end users do have lots of strategies to respond to carrier differentiation, and that when one thinks of the problem that that net neutrality is trying to address, technically one has to consider what the but for world would be in a world where there aren't any rules and in that but for world one has to consider what these kinds of responses would be and do a little more in thinking about, you know, what the implications of that might be. Another learning that we took away from this is that the end user responses can occur even when the kind of traffic management differentiation we are seeing is good. And so there may be another be that really hasn't been adequately addressed yet in the debate, and maybe something we're going to have to sort of observe. It's certainly something that the carriers perceive themselves having to deal with, when, for example, they look at certain types of users that they feel are using dramatically more resources and they are trying to figure out, you know, what is a fair way to recover the higher costs associated in providing those customers with service while at the same time, you know, not denying traffic that in fact, you know, the network can carry and ought to be able to carry but only if it's able to cover its costs. And then the responses that end users have, though, now I suggest are imperfect and most of them depend on carrier using a particular model of discrimination and the carriers, if they use a more sophisticated model can perhaps render ineffective. And so the only really sure way for end users to provision around this is to be able to bypass the bits path over which they are seeing experience in discrimination. So technologies for doing that and options for doing that on facilities based competition are critically important in addressing that challenge. The other thing that comes out of this as you begin to unpeel this onion you realize that with most interesting problems and the complexity gets worse not easier as you go forward. So the net neutrality problem is complex and it's going to remain a concern that we think the welfare and efficiency in equity gain of this -- not having it and allowing the market to just play it out. The implications are ambiguous. So it's not clear wheahter or not the -- what happens with this arms race and what the costs of playing out the arms race, in the absence of regulations would be. If we had the regulations we understand there could be real problems with discouraging effective market behaviors. Last slide. With this slide I'll conclude. So the broadband future we see is complex and heterogeneous. And so I think, you know, my own view is that there needs to be a new answer response and along the lines of something like what John was suggesting, and that, you know, there is a real need to try and get some free clear framework, a regulatory framework so that the industry and everybody knows what the game is going to be and what the critical concerns are going to be. The key, since the key element is going to be to, you know, the key element insuring that end users do have responses and ability to avoid addressing this whole net neutrality problem is more facilities-based alternatives than there are some obvious issues in terms of infrastructure investment. Let me just give focus here on the technical issues. With respect to municipal entry, a lot of folks, you know, make the false conclusion that when local governments or local communities build infrastructure or get involved in the infrastructure provisioning question, that that's a, you know, sort of binary good, bad thing and they do it one way or they don't do it. The answer is it's a very complex mix of strategies they face. The particular technologies and strategies they undertake, how they do that is a very complicated thing and has big implications for what sorts of net neutrality problems may happen. For example. If they do like a fiber deployment that's an open access platform, then that really does go a long way towards eliminating concerns, most of the net neutrality concerns, but such an infrastructure plan is unlikely to make sense in most communities, and other alternative sorts of strategies, if they make sense at all, need to be evaluated in this. The other thing is that a lot of the sorts of alternatives that we talk about really depend a lot on wireless and new sorts of wireless technologies, so making sure that we have a really vigorous commercial market for new wireless technologies I think is critical to addressing this problem, and there are so many different wireless items, we may get into a little bit of that in questions and answers but spectrum reform is obviously a key element of that. With that let me thank you very much and let's go to questions. (applause).

>> CHARLES GOLDFARB
Before I ask questions could people please fill out the questions they have and if there is someone who from FTC who could pick up some of the questions, I'll have one or two to ask, but then questions from the audience really are appreciated. John and Bill specifically said they'd like to get as many, as much audience participation -- participation as possible. While they are coming up let me start with a question. To date most deployed broadband access networks are wire line and thus the tools that have been developed to manage them are tied to wire line technologies, but let's now talk a little bit about the wireless technologies, Bill gave that list of them. Do any of the potential wireless technologies have technical characteristics or cost characteristics that would make it more or less difficult for the wireless broadband access network provider to discriminate, than it is for the wire line provider to discriminate? And along with that, do any of these wireless technologies have technical or cost characteristics that would make it more or less difficult for the independent applications providers, and end users to undertake counter strategies if they face discrimination by their wireless broadband access network provider. So John, you want to start? And then Bill, feel free to step in.

>> JOHN PEHA
Sure. If we're talking about a broadband packet switch network which is the place to start as opposed to, you know, Voice telephony, many of the things we've said we think are the same but there are a few interesting differences. One is if you have a network that has multiple paths into the Internet, for example, a mesh network in particular, then it becomes a whole lot harder to discriminate and there are a lot more counter measures that become possible. If you have mobility, somebody moving from WiFi hot spot to WiFi hot spot, some of the techniques become harder. If you have sharing which you do in a lot of these systems, let's say you have a big WiMAX broadband system and you are in effect sharing capacity, that actually may make the technology of discrimination a little harder but it may make the need for it a little greater because now you are sharing you have reason to, greater reason to worry that a small number of users will dominate the resource, so there are some subtle differences. The market differences may be more important than the technical ones, though.

>> CHARLES GOLDFARB
Bill?

>> WILLIAM LEHR
First off, everybody should be very clear in their mind what a level point one is, spectrum is perceived to be a very scarce resource, RF spectrum so generically your bandwidth is more of something, a resource you are going to be more concerned with in the wireless world. So you know, equivalent levels of performance are in some sense sort of the need to, for example, carefully manage traffic on a wireless network is greater. A second important issue is that, you know, the architectures of mobile networks versus these other alternative sorts of, you know, broadband wireless fixed networks that are just emerging based on things like WiFi and meshes and newer technologies like WiMAX that are just now beginning to roll out, are pretty different. And, for example, with the sort of mobile carriers, because of the way they actually provision customers, they are probably in a better position to discriminate on a customer by customer basis if they wanted to, and end users' abilities to sort of do much about that because of the closed nature of the current mobile networks is sort of -- is tougher. It's sort of more attenuated. With these other sorts of, you know, sort of mesh, WiMAX types of networks, I think John addressed most of the key points there that I would have mentioned.

>> CHARLES GOLDFARB
I encourage more questions to come up. I have a few that may be verging a little more on policy than on technical, so I encourage people to ask technical questions for this group. But one question that came up was the fundamental question of incentive assumes service -- that the service provider owns the transport. Why not correct that dysfunctional assumption and assume a not for profit or a for profit road system so that there is a distinction between I guess the access provider and applications provider.

>> MALE VOICE
(inaudible).

>> WILLIAM LEHR
Who is that to?

>> JOHN PEHA
Okay. I mean a for profit road system sounds similar to what we have except -- I'm sorry. Not for profit?

>> MALE VOICE
I'm arguing we have a (inaudible).

>> JOHN PEHA
So you would like an for profit road system so probably a monopoly or single provider if you like, a single provider not for profit? If you were building a new system from scratch it would make a whole lot of sense to say is there a strong an economy of scale in some part of this, perhaps the last mile, and then you could ask that question. But we are not building a system from scratch.

>> MALE VOICE
(inaudible).

>> CHARLES GOLDFARB
Let me perhaps have some other questions.

>>MALE VOICE
There have been -- there has been a lot of hand ringing about US broadband networks providing less bandwidth than some foreign networks. In my conversations with Bill and John the two of you have suggested what might be two technical truths, one, that no network architecture has a bandwidth constraint that the network provider can't buy its way out of. And secondly, that to attain a higher data rate with any given network you must serve fewer homes or less distance. So it seemed like there was, you know, one was looking at it dynamically you can always sort of buy your way out of it, but there are certainly constraints when you have a given capacity. That this seems to suggest that the first one suggests that the constraint is cost or the time to develop necessary hardware and software rather than a technical constraint. The second one where you are saying well, there is only a limited amount of homes that you can serve or distance suggests a technical constraint. And I'm sort of curious about this tradeoff of cost and technical constraint. Assuming it's an important goal to substantially increase the bandwidth capacity of our broadband access networks since there is the argument that we don't have very high capacity at least to the end user, which of the various wire line and wireless technologies potentially available for broadband access are likely to face the fewest technical or cost obstacles to achieving this goal? And for the wireless option, is lack of available spectrum likely to be the greatest constraint to providing large bandwidth?

>> JOHN PEHA
I think the distance we're talking about there is the distance to some point where you aggregate data. Could be a central office in a telephone system. It could be a cable head, something like that. So if you are limiting distance it means you need more of those aggregation points and that's expensive. Certainly the engineering economics change with the density of users, so what is most cost -- anything is possible, you can throw money at it. But what is most cost effective in a rural area might be different from what is most cost effective in an urban area. Everything gets more expensive in the rural area except picking of roads. But generally the wireless services seem to show greater promise there and particularly if you want to cover large areas, wireless, not at a very high frequency is rather important, so there are interesting opportunities in the digital television transition that some 700 megahertz spectrum becomes available, there may also be opportunities to share spectrum more than we have in the past at frequencies that allow you to cover large areas in rural areas.

>> CHARLES GOLDFARB
Bill.

>> WILLIAM LEHR
Let me say a couple things. First off, broadly when one talks about the available bandwidth, the technical limits, different media, physical media have different transmission. When Jon and I were talking earlier with Charles, we were saying that essentially those technical limits are unlikely really to be the binding concern, although they are relevant. So, you know, broadly, you know, it's harder to get band width across air so wireless is the technology that generally is going to have less capacity than copper wire, and copper wire has more capacity if you go over copper wire for shorter lengths. So, you know, DSL at a megabit per second works, you know, pretty far from the central office, but if you want to go at a much higher rate then you need to be going --, driving that copper wire much closer to the home. Coaxial cable used by the cable television carriers has a lot, you know, it's bigger, thicker wire and a lot more bandwidth on it, but the way they use that is a shared cable that passes many different homes, and so the bandwidth that's available to individual home, you know, you can't compare that as the whole cable on its average rate although potentially on a burst rate you can. And then, of course, the biggest capacity that you get into a home is if you have fiber all the way to the home. The cost of scaling any technology depends on the up front investment and how far you want to go. So if you know you want to be able to arbitrarily scale the amount of bandwidth to accommodate lots of competitors or, you know, dramatically new expansions in use, then put in fiber, but putting in fiber is expensive so that affects the cost model. The architecture of the different technology, that depends very much on who carrier and the provider of some of these are and what it is you are trying to do is going to influence how easy it is to scale. And so key elements of the architecture include the choice of media. Is this copper wire, coaxial cable or are we talking about fiber plan? How you are doing the back hole aggregation, how many homes are you serving off of a node and that you are pulling over a common wire until you are eventually connecting to the upper Internet, all those sorts of decisions have big implications for your ability to expand the capacity at a low cost to additional homes. Whether the traffic is symmetric, how much up stream versus down stream traffic do you expect and are you provisioning for, and finally which I think is the thing that has been overdominated or really dominated most the net neutrality discussion, is what are you doing in terms of managing the traffic especially over the shared elements of that network. If you cut that shared infrastructure into silos, for example, say, you know, this infrastructure is, you know, this is dedicated to this application, this is dedicated to that other application, you are going to have less effective capacity than if you are able to share the whole infrastructure using the advanced kinds of techniques that John talked a little bit about. But of course, as soon as you are sharing all of that as one common infrastructure across applications and potentially providers, then you are going to have quality of service spill over effects which means you are going to have some of these sorts of net neutrality concerns that you are going to need to try and mediate.

>> CHARLES GOLDFARB
Some technical questions have come in. This one is for you, Bill. It's how realistic are the consumer strategies? Many consumers have no choice of carrier or no technical expertise to deal with harmful discrimination.

>> WILLIAM LEHR
I think it's a really important question and I think that there are a couple of things here. First off, the question is of the net neutrality issue is largely a perspective one because there isn't a lot of evidence that currently today really bad things are happening. Which is good for the point of view of the end user responses because a lot of the end user responses we talk about are things that are not easily or really widely able to be done today. The other bit of response is you of course don't need every user to be able to do this to have the benefits of these sorts of strategies out there showing up in the market. For example, a lot of the kinds of application programs, especially if they have automatic updates, the users aren't even aware of how these new application programs are responding or changing to market conditions. And so the, you know, a few folks out there can actually do the work of a much larger consumer base. On the other end, as we conclude in this thing, we think the end user responses are somewhat limited and are likely to be most effective against the least sophisticated versions of discrimination. And so that's the reason why we conclude that this really is a valid concern and that it is not sufficient based on our analysis of the fact that end users have options to conclude that there is no problem here.

>> CHARLES GOLDFARB
I'm not sure which one this one would go -- this question goes to which of you. But could we switch to a different network structure where video and heavy content doesn't run on TC/PIP networks?

>> JOHN PEHA
We've long had an infrastructure where video doesn't run on TCP/IP. The question is whether we should switch the other way, and, I mean from a technical perspective, either are certainly possible. I would say packet switching has an advantage when you are not always downloading every -- you know, if you are always downloading all content all the time there isn't a big reason to move to packet switching. If you move to a more on demand model for video, if TiVo is closer to the model of the future of video as opposed to what we are used to then there is perhaps a reason to switch to packet switching so when that capacity is not used for video it can be used for something else.

>> MALE VOICE
I'd like to follow up on one of the responses you gave earlier, Jon, when I had asked a little about spectrum as a constraint and you talked about the 700 megahertz. I think this thing gets into the issue of how spectrum is made available and auctioned off. One of the competitive users or demanders for spectrum on one hand there have been arguments from satellite companies that pass spectrum options have had license areas, geographically that were too small to have a nationwide coverage. On the other hand, smaller regional carriers have been arguing to have very small license areas in an argument they would be focusing on -- they would be focusing specifically on rural service and therefore enter. SO I guess a question I would have is how likely is it that the technology is available and would be used if there were a nationwide license given to have a nationwide wireless network made available through an auction?

>> MALE VOICE
Predicting the market is notoriously hard I can certainly say that having the license cover too much area is a problem, that is somebody who only wants to serve urban areas may get the license and ignore the rural areas. Having a license that covers too little area is a problem. We've seen cellular carriers have to piece together lots of licenses over the years, because by no fault -- the FCC doesn't know the future, it seems with 20/20 hindsight those were too small. I would like to see more efficient secondary markets so we can correct the fact that we cannot absolutely predict the future and deal with some of these problems. I don't know where the best way to go is a nationwide network or a bunch of regionals.

>> WILLIAM LEHR
From the technical point of view certainly operating down in lower frequencies, the beach front property in spectrum is below one gigahertz because spectrum -- because signals at that level can -- at that low rate don't need line of sight, and so that's really really valuable. It means lower costs for deploying infrastructure, and so making spectrum available down there, more spectrum available down there for commercial communications services would open up new options to help alleviate last mile facilities competition concerns. And so I think it's very important that we try to figure out how to free up some of that spectrum. With respect to the end user responses, a lot of the models that I talk about I think would be given a big shot in the arm if there were more opportunities for end user deployed kind of networks, municipal WiFi, community based networks. It's not just using WiFi but using other sorts of emerging wireless technologies. And so I've argued in favor of the white space access. I think that would be very important in energizing the wireless market. And then also I've argued in favor of additional unlicensed spectrum in that band, and how you get it into the market, you know. What auction mile. That's a complicated decision I think, debate and discussion that I think is beyond the scope here. But definitely the question about what we do with that 700 megahertz spectrum I think is an important aspect of this whole net neutrality debate.

>> CHARLES GOLDFARB
Well, since in fact most of the questions that are coming in are really policy, I think just for the last few minutes I'll just turn to both John and Bill and ask if you have any closing remark or statement or anything that's come up in this hour that prompts you to want to expand on what you've presented? And if not, we'll be in the very unusual situation of being five minutes ahead rather than 20 minutes behind on our schedule. Bill, thank you so much, and John, thank you. Look forward to using some of the information they've provided in the next two days. Thank you both.

• See FTC Workshop Transcript for index