______________________________________________________________________ DRAFT TRANSCRIPT APNIC 20 Opening Plenary Wednesday 7 September 2005 9.00am ______________________________________________________________________ PAUL WILSON: Good morning and welcome to the Opening Plenary of the 20th APNIC Open Policy Meeting. Thank you all for being here. We've got a few speakers this morning in the Opening Plenary, who I'd like to introduce shortly. I'd like to firstly thank our sponsor for the day, which is Nominum. And Nominum is a gold sponsor and has contributed substantially to helping to run this meeting. So thanks to Nominum. They have a lucky draw happening and you can win an iPod. So you can enter the lucky draw at the APNIC help desk and that will be drawn toward the end of the week. So please try your luck in that competition. We have lunch today in the grand ballroom 2 on the first floor. We also have a social event tonight, which will be sponsored by VNNIC. And that will be not in this hotel but you'll need to catch a bus and buses are leaving from the hotel from 6:30pm to 7:10pm. And there'll be a few more details of that event later in the day. But that should be a very good one - the Vietnam Museum of Ethnology. So moving right along, during the Opening Plenary session this morning, which will run through to morning tea, we've got several speakers. Firstly, we'll hear a welcome from the local host of the meeting, VNNIC. Tom Vest and Geoff Huston will speak after that. I'd like to hand over to Mr Thuy Nguyen, who will give a welcome on behalf of VNNIC, the local host of the meeting, and I'd like to say thanks right now to VNNIC for the great support that they've given to this meeting and to APNIC in organising the meeting. It's been very well organised and VNNIC has worked very hard to be the host for this meeting. So thank you very much and please Mr Nguyen... THUY NGUYEN: Ladies and gentlemen, it gives me great pleasure and it's an honour to be here with you this morning to attend APNIC Open Policy Meeting 20, hosted in Vietnam for the first time. On behalf of the hosting organisation, I warmly welcome distinguished guests, ladies and gentlemen, from all economies in the region and the world who come here to attend the APNIC meeting. Ladies and gentlemen, Internet in Vietnam was accessible since the end of 1997. Since then, the Internet in Vietnam has continuously expanded in terms of both quality and quantity. So far, the Internet population of Vietnam reached 9.68% of the population. Along with the development of the Internet in Vietnam, Vietnam Government has also built regulations and policies which enable the Internet resources to be developed and used effectively. In April of the year 2000, Vietnam Internet Network Information Center was established as an agency to formally manage, control and register national Internet resources; later on, as an office and representative of Vietnam Internet community, Internet organisations and forums. Despite operations, thanks to the close involvement of the ministry, support from the Internet community, together with our efforts, to date, VNNIC has made remarkable results in terms of stable, national DNS operations, network service and contribution to the Internet in Vietnam. With the result in Internet resources management, in the year 2003, VNNIC was recognised by APNIC as National Internet Registry, which brings comfort to the local Internet community to register and use Internet address. May I take this opportunity to give our appreciation and thanks to APNIC for their activities and assistance in guiding, coordinating countries to assign and management Internet resources. Again, I would like to express my sincere thanks to APNIC for their support of VNNIC and making this event possible in Vietnam. It will be a very fruitful opportunity to confirm close cooperation between VNNIC and all Internet organisations and, at the same time, the growth of the Internet of Vietnam. And I hope it will also be a chance for us to exchange information and experiences regarding the Internet field. Last, but not least, I would like to take this opportunity to thank for their valuable attendance and believe that it will make a real contribution to the success of APNIC 20. I hope you will enjoy Hanoi, as well as learning some traditional culture of our peoples. I wish our meeting success. Thank you very much. APPLAUSE PAUL WILSON: Thank you very much, Thuy Nguyen. The next speaker is Tom Vest of Packet Clearing House. He'll be speaking about innovation and infrastructure and the digital divide: ‘Lessons from the history of the Internet’. Thank you, Tom. TOM VEST: While I'm doing this, let me start by thanking our local host, VNNIC for the opportunity to be here and also I'd very much like to thank APNIC. Five years ago, very close to the beginning of my career as an operator, I had the unique good fortune to speak in this very capacity at the APNIC 10 Open Policy Meeting in Brisbane and hopefully in the intervening five years, I've actually learned a few things which will help make the correct presentation. So the title of my talk today is 'Infrastructure, Innovation and the Digital Divide in Asia: Lessons from Internet production history'. So I think we're all aware of the rising attention that the operations, the Internet operations field is enjoying from a variety of segments of society, including national governments and inter-governmental institutions. I would say much of the source of interest, the source of contention, is based on issues of sovereignty. Sovereign prerogatives, national security, things of that matter, a fair proportion of the concerns are couched at least in terms of concerns about economic development. About differential rates of growth of Internet access and content, Internet usage. In other words, of the digital divide. Having begun my career as a person in the international policy field, I would say that matters of sovereignty are not subject to much empirical discussion. They're generally informed by facts, data, research. However, that's not necessarily true on the economic side. I think the questions can be usefully engaged with research data, facts that we in the Internet production policy committee are - some of which we have unique access to. And I'm going to be focusing on that. I would say the economic part - this part is actually unlike the sovereignty issues, is not a matter of pure politics. Just 50% politics. So I'm going to suggest a natural language interpretation of what we can learn from the Internet routing table and from the archives of the routing table. The digital divide, as it is described by policy makers, is often defined in terms of symmetrical growth in users - subscribers, opportunistic users – usage - so timeline and multiple access methods for each user to reach the Internet - and actual uses - or diversity of content and services, especially including those which have more salience for people around the world. And generally when we talk about a divide, or asymmetry, we define it in terms of comparisons across countries. So I would suggest that using this sort of natural language - interpretation of the routing table - that we have quite a bit of data to inform these questions. An interpretation of the routing table I believe could be of great use in discussions of the digital divide. So, again, speaking just in common non-technical terms, you're summing up users and usage, uses, content and services, and scaling factors that we're all familiar with, what you have in effect is a sum of the, sum corresponding to the IP addresses injected to the routing table by ISP or AS. (Refers to slide) This is all old news to everyone in this room. But very quickly, where we're going to be looking to counter production, each router has a specific view to pass through all resources that are currently in production. All of those views will vary wildly in almost every aspect, except for one - all will share a common view of the network or AS origination for each and every Internet resource online. So, when you're talking about Autonomous Systems... (changes slide). So, individual Autonomous Systems can be plausibly associated by the Whois database with the national jurisdictions of ownership or administration. Of all the records, the Whois records, which are currently in use, the ones associated with Autonomous Systems is the most complete and the most accurate. That's not to say that it is without all difficulties. But I would say that you can enjoy confidence looking at the Whois DN level. In fact, given current initiatives, which are under way around the world, across several of the IRRs, this is probably something we can expect to have greater confidence in over the coming years. So, if we can, in fact, associate the digital AS numbers, we can associate every router's IP with an autonomous system and we can associate each autonomous system with a nation of beneficial origin or administration, we can have in fact - we can derive a sum of Internet production. Of course, this does not mean that Internet resources are counted to necessarily correspond with the geographic area of the nation states. This is more a means of counting which is consistent with the way other multi-national corporations are counted in other economic analysis. So the question is not where the resources are going, but who is the ultimate beneficiary and the ultimate authority over those. This is a way of approaching this data is not new. Analysis of the routing table has been going on for a decade or more, mostly for within the community interests and requirements, issues dealing with scarce resource husbandry, and dealing with other network administration and network optimisation questions. There are lots of individuals in institutions, many of which are represented here in the room now, which have been actively involved in this. You don't see, until very recently, much that actually has been trying to associate with these logical layer endings with geopolitical endings until very recently. The one exception is a single-point study that CAIDA did in September 1, which was very interesting. And, in fact, it prompted me to think about what that kind of research might look like on a client search basis. The router archive and, in fact, a few other archives are publicly available resources, online, available to anybody with a means to manipulate the data and they do provide, with a bit of work, I think, an instructive time series view of the historical trends in Internet production at the national level. So this is, by narrow region, from over the course of the routeviews archive, which begins in November 1997 and continues to this day - these are taking a look on November 1 of each year, the earliest full snapshot which is captured by routeviews to the - 2005 is May 1 so it's 12-month intervals and a 6-month interval there at the end. So the aggregate is just a place to begin. There are a few things you might - you know, things you might infer from this but I want to focus specifically on issues relating to the Asia Pacific region today. (Refers to slide) Here's a finer-grain look at the growth of, say, the national network economies of the economies associated - that are associated with APNIC from 1997 until May of this year. And so you can see that there are - there are interesting, potentially very interesting, variations in the growth trends and - both in terms of rates of growth and in variations at different points in times across countries. And, again, if this view can be observed as - in an economic sense as a measure of overall national network, national economic - national network economy production, then it seems to me that it would be useful - looking at this data and thinking about what makes certain national network economies have certain kinds of trends as opposed to others is worthy of further consideration. Here's a closer view of the narrow region for South East Asia. Think - I invite you all to take a look at the data and draw your own inferences for the time being. So what does it mean in fact - what can we learn or say plausibly from this data? We're looking at national production, it sort of begs the question - how much should there be? How much should one expect? I would say one way of looking at this, which is particularly salient given current policy interests, is to look through these logical layer assets and compare them, these logical network ends and compare them to physical network ends, which we actually can count with some precision at the national level, thanks to the ITU, the OECD and other institutions. This sort of approach, I think, lends itself to a kind of intuitive observation. If you think about - if you had experience as a hostmaster or through IT planning for an individual ISP, you might imagine yourself in a position - looking at the IT data, being the hostmaster or the provision manager for that national network. You might imagine what the network would look like in that case if you were looking at the certain distribution of dial-up users plus broadband users and you're hosting a certain quantity of content locally. So I think - some countries actually, the data actually is suggestive of these kinds of comparisons. So, in order to usefully map logical area assets against physical media, I think looking in as little detail as usual. With the legacy copper infrastructure: (refers to slide). Then fibre, beginning deployment in the early '80s and then really picking up in the '90s, first in the sort of telecommunications and PSDN core and then internationally over some terrestrial routes. And then, more recently, in some places, we have a secondary access network, the Coax networks have been deployed eventually for television. So the timeline for which those physical media were deployed in their different physical characteristics is interesting but I would say that actually they're entering into the reality where we're examining - coinciding with major changes in market structure which are at least as important as the technical changes themselves. So the introduction of optical fibre in terrestrial cores and then internationally occurred at the same time that really in many economies the whole concept of a value-added service as something that was independent of the primary PSDN service was defined. So the medium and the market ending itself kind of came into being at the same time. And not long after that, wholesale access - in other words, the operators being able to leverage these core access, these core bandwidth resources shortly thereafter in some economies were able to actually begin to enjoy wholesale access-to-access networks as well, at least the copper-twisted. And then, of course, with the introduction of Coax in some markets, there are other significant developments - the introduction of access segment competition. So, does it matter? Variations across these different kinds of assets is significant. Based on some fairly detailed measures of the US - United States-based copper infrastructure - and inferring from ratios of miles to lines and looking across the ITU data for 2003, it's plausible that there are, you know, between 9 billion and 10 billion miles of buried legacy copper-based infrastructure in the world. Contrast that to perhaps, perhaps a tenth as much optical fibre over a much thinner set of routes and the vast majority of which are duplicative. And Coax often lauded as an alternative to facilities-based competition as opposed to other mechanisms is in fact only in evidence. There's only about maybe, you know, the ratio of copper to Coax is 10 to 1. So there's a good many markets which will never - not in the foreseeable future - be able to enjoy facilities-based competition between these two access networks. So does it make a difference? Does access networks, does this notion of physical ends, does it actually - can you plausibly say that it has some bearing on national network, national Internet production? Well, here's a look at the top 50 markets (refers to slide). I think this is 2003 ITU data. The red line is main lines deployed as measured by the ITU (refers to slide). And the blue line is BSE net productions routed, unique public Internet addresses which resolve to the countries associated with the IT data point. And this is November 2003. So it looks like there may be a broad resemblance between the two trends. We'll keep that in mind. (Refers to slide) Again, looking at it in fine-grain in the Asia Pacific region, I think it's potentially a useful exercise. It's going to take more time than we can devote here at the podium today but I commend you all to have a look at this on your own with the slides. (Refers to slide) And, again, here's an even finer-grain look at some of the endings in the South East Asia region where they currently stand. Again, for those of you who are familiar with IP address planning for asynchronous dial customers and things like this, I think it's plausible to think about the national access network as a denominator over which you might think about what it might be like to deploy IP resources. So, on that basis, I think actually a useful approach for tackling the challenge of the digital divide is to foster better, faster, more thorough conversion of physical resources to these logical resources. And it begs the question - building ISPs is challenging, demanding and expensive but the expense is not in the same order of magnitude as the cost of digging holes in the ground across a city and burying conduits and trying to connect lots of physical facilities in a high-density area. So if you can leverage - for any point in the history of a given region or countries, physical telecommunications development, I believe a useful goal would be to make sure that that endowment is going, is being utilised to the greatest extent possible to deliver these key Internet resources - Internet usage online, Internet uses, content and services. So there are some - I have some - these are hypotheses for future testing. What are the factors which contribute to better conversion rates for that ratio? And I would say, really, the key is making sure that those segments, those media, that they're efficient and effective bridges that connect them. There is no barrier between one. Becomes a bottlenecked overall national network development. And that's the bottom line - to think about how best to leverage whatever set of numbers you have at any point in time, to deliver as much of those resources as possible. OK, back to that first chart - looking at this again (refers to slide) You may or may not see a pattern there. Do you see a pattern here? So the yellow line is the number of Autonomous Systems that, again, which Whois suggests are associated with that national network. And I believe, actually, that Autonomous Systems and counting them and looking at them and understanding what makes them emerge in some places, in some markets, what are the conditions which foster the diversification, proliferation, of ASNs, basically, diffusion of control over the Internet’s logical layer is an important puzzle for solving for understanding the digital divide. This is again another trend for future observation. The speckled slices (refers to slide) are actually AS numbers which have been consistently routed since before or from the first day of the routeviews archive in 1997 and which continue to be routed today. In some economies, those continue to be the dominant operators. Now, if you look - this is the total number of AS numbers that are currently in production as of May 2005. You see that legacy operators constitute a very small minority; that there's been quite a lot of diversification and proliferation of AS numbers in some parts of the world. But that growth has been - the growth rates are very divergent across different economies. Again, I think, although the gating is arbitrary and related just to the data, I can't really look back at routing tables before 1997, so I'm just going to stipulate that that's a meaningful date. But, in fact, of the 1,600 or so Autonomous Systems which were in production in 1997 a good number of them actually are - they would not be - they would conform to your, sort of, commonsense understanding definition of a PSTN operator or a legacy-incumbent operator. (Refers to slide) Here's a closer view. Again, looking at how legacy operators, how they persist in their varying levels and varying rates across different economies. (Refers to slide) And here is looking specifically at the subregions of the Asia Pacific. So get to the money question - going back through my chain of implication, if legacy operators do correspond to facilities-based owners, to PSDNs, to incumbent operators, and if, in fact, the notion of bottlenecks between different infrastructure types does represent a challenge to national network development, then we want to go looking for evidence in a way like this (refers to slide). And this is the reverse scientific approach. It is a quick check to see whether or not one can say that the persistence of a strong incumbent or a set of incumbents who collectively represent a substantial portion of the national network economy, whether that looks to be statistically to be a benefit to a network economy or a hindrance. The blue columns are the overall percentage of growth between May 2005 and November 1997. Pakistan is off the scale. It's been 900%. And the rest are within - actually, about 3,000%. The rest are within a magnitude of their initial state. The question is does it appear that - the red columns, I should say, represent the share of growth during that period of time which is accounted for by AS networks, by Autonomous Systems, which entered the routing table after - actually some time after November 1997. I observed them first on my next angle slice of the data, which is November 1998. So, if incumbents have an important role in national network development, we would expect those columns, the red columns to be rather short. The only observation I could make at this point without further research is to say that the data does not seem to support the idea that incumbency, the firm-fixed incumbency is essential to national network development. Now, whether the converse is true and what, in fact, are the conditions that foster that proliferation of non-legacy operators, which is to say new Autonomous Systems, new networks, new network enterprises, is a very broad subject, which I'm not going to engage in now. But it's a matter that we're studying actively at Packet Clearing House and something that we do look forward to sharing further insight about in the future. So, anyway - my conclusions. It seems to me, based on this data, that proliferation of Autonomous Systems is a useful measure or benchmark of efficient - of one element of efficiency which contributes to national network economy, which is to say domestic wholesale capacity market, the course fibre part. And the growth of Internet production without corresponding proliferation, diversification, of AS numbers may be, in effect, indicative of bottlenecks which may be harmful or may actually impede the rate of national network growth. And, again, if I want to pile a speculation on top of speculation, I would say that the digital divide is likely to close farthest and fastest where production of users, uses, usage - which is to say Internet production - is not bottled up in any one of those segments and is not segregated but rather can go on freely across all of those infrastructure segments and market segments. So - less speculative. It does seem to me that a natural language interpretation of the routing table, coupled with an operationally informed interpretation, can represent a useful form of - a useful contribution that we can make to the ongoing policy dialogue with WSIS, within national network economies. And, again, I think, actually, matching discussing layer 3 in a context of layer 2, layer 1, is a useful way of kind of talking past some of the impediments which have, in the past, made it difficult for us to make progress on these policy measures. And that's really all I have to say other than thanks once again to the people who made it possible for me to be here. I take questions up here? PAUL WILSON: If there are any questions, sure. One or two? If not, then there may be time at the end of the session before morning tea. So I'd like to say thanks to Tom. APPLAUSE PAUL WILSON: The next speaker is Geoff. Geoff will be presenting a topic tentatively titled ‘Future of the Internet Industry’ but I think he's refined it a bit since the initial brief was given. GEOFF HUSTON: Thank you and good morning. It is a pleasure to be here. It is always interesting trying to figure out what to say in plenaries. Repeating conventional wisdoms and reinforcing what you already knew, although comfortable, is hardly fun. I would like to take a somewhat different view and challenge you to think about some of the issues around the business we work in in different ways and of course when you're looking inside this industry for a target-rich environment, you need to look no further than IPv6 to actually find a healthy array of targets you can hit at. So in the spirit of at least challenging some of our views and looking at this industry and what is going on with IPv6, I would like to offer you some views about a personal idea of where and why IPv6 is moving. Like Tom, I have spent a lot of time looking inside the statistics of the Internet and one of the richest places to look at, as you may be well aware, is looking at the interdomain routing table. If there is a map of the Internet, if there is a map of deployment, it’s routing. Routing describes not only how to get places but actually tells you how many places there are. So let's have a look at IPv6. Over at route views, the University of Oregon, it is a wonderful site. Every day and every two hours they do a snapshot of the routing table, they do a snapshot of the IPv4 routing stable and IPv6 routing table. The IPv6 data goes back to March 2003. What you're seeing on the screen here is the number of entries in the BGP routing table for IPv6 since March 2003. Not very long. 2.5 years. Number of entries - bit interesting. In 2003 there were 400 routing entries in IPv6. Today there are an astonishing 800 routing entries in IPv6. Over the last eight months in 2005 it has grown by 100 entries. You can see a bit of noise there, there is a certain amount of aggregation and deaggregation happening. Overall it is a smoothish curve. Over the northern summer, folk have been interested but it is a relatively small routing table. By comparison, let's look at what happened in IPv4 over the same period. In March 2003 there were 120,000 entries in the BGP routing table. Today there are 165,000 entries and over the past eight months of this year it's grown from 150,000 to 165,000. “Oh,” but you say, “it's not the same thing ,” and you're probably right. IPv6 has much more visible aggregation properties and what we're seeing in v4 is a different way of routing. A lot of folk not only advertise their aggregate but also spin out more specifics for traffic engineering and a lot of the noise we see in the IPv4 routing table has nothing to do with the actual level of business activity, more to do with traffic engineering and trying to shape the way the packets flow. Maybe that is an unfair comparison. Let's try a different view. AS numbers are interesting because if you want to know the number of industry players who are actually originating IPv6 routes, then the best way to count that, irrespective of their size, just look at the number of players who are actually playing - AS populations are a very good way of doing that. Here's the number of folk actually originating IPv6 prefixes. You can see now why there's very strong aggregation going on. The IPv6 routing table has 750 entries in it and they come from 550 unique Autonomous Systems so in general most folk advertise one prefix in v6. What's the growth rate? Again this the same 2.5-year period. My apologies - it is hard to see that from any further away than I am but it spans from 300 to 550 from March 2003 to September 2005. This year, how many folk joined this public network? 50. So the table grew from 500 at the start of the year to 550 now. 10% growth over that period. Interesting - that's a pretty stable growth rate. It is not exponential, not declining. It seems almost linear. What we're seeing is an average rate of 100 new players a year, give or take 10%. That's the growth rate of v6. Let's look at v4. Same period - what are the figures showing us? The number of Autonomous System players grew from 14,000-odd to a little over 20,000. In v4 the table is much noisier. On average for each Autonomous System, there's eight prefixes being advertised. There's not strong aggregation in the v4 routing table but the growth rate is high even in the number of players. By comparison in IPv6, 50 players joined this year. In v4, the growth rate is 18,800 to a little over 20,000 so a growth rate of 1200 or so new players, much higher growth rate. Let me put that into perspective with the two graphs together. The top line is v4 and near the X axis is v6 growth rates. Why hasn't v6 been more explosive in its uptake? Why are industry players so reluctant to actually join in? Cost of addresses? I don't think so. There's a lot of IPv6 addresses around. What is going on in the industry to understand why the uptake is so slow? We are going through boom and bust and I think we're starting to learn a few things and as we move from experiment to mainframe and mainstream industry from a few million dollars to a multibillion-dollar, multimillion-user platform to be part of the mainstream communication industry, our own behaviour is playing. We have gone from optimism to conservatism. We are a very conservative industry. These days you need business plans. You can't dream up new technology and get the company to invest in it. Now it is all about knowledge and capability rather than just simply optimism. If you look at your own situation in the ISP industry, how do you deploy a new product? You've got to do business plans and show some competitive financial return for your product across its life cycle. It is now about the money, isn't it? Now we're dealing with investors who are much more conservative in their view. They're not interested in seeing you take a risk with the shareholder value. They now want much more assured return from their investment. This is a very conservative industry and now the communications industry is now - instead of trying to pump out new technologies every day, it is looking at value-added service bundling, trying to wrap up access and content and get greater return from their existing investment. That's a bit interesting. The industry itself is not moving the same way it was five years ago at the height of the boom, it is trying to sweat the existing infrastructure, use the routers for longer, stop constantly cycling technologies inside the network and make what they have bought generate income. It is a very different view. Now let me map this into IPv6 and look at it from what I would term almost a biological viewpoint. In 1831 to 1836, Charles Darwin sailed around the world in the ‘Beagle’ and one of the islands he went to, the Galapagos Islands, as I recall, he observed some amazing things about species on that island and he came up with a revolutionary view that was published in 1851 called the ‘Origin of the Species’ and the view was very heretical at the time. It said biology and species are not invariant, they change over time and compete with each other in various ways. He was pushing forward a view of evolution. There are other ways of competition as well. In looking at v6, I think can I apply the same views. What is the expectation of IPv6 in the industry? Are we looking at what I think many of us thought was going to happen which was evolution, the existing service providers, industry players, end users would slowly realise the phenomenal advantage of IPv6 and little by little IPv6 would come along inside the same networks, dual stack for a while and over time IPv6 would come into mainstream marketing. That's certainly a view but there are other ways of looking at it. One view is what happened to the telephone industry when the Internet came along, which was not evolution, it was bloody. There were bankruptcies, money to be made and money to be lost. All that happened in a very short period of time. The view of IPv4 on the industry was revolutionary. The shock of the new changed most of the industry in a very, very short period of time. It was indeed revolutionary and in the view of IPv6, is IPv6 the natural revolution of v4 or a competitor? Is it actually vying for the same market share? Is it competing? Does it offer revolutionary different aspects that will ultimately cause it to happen? Of course there is a third option because we have seen this also in biological systems and it is called extinction. Is this too little? Does it offer not enough value? If it doesn't, it will die. Extinction has been an option for many years. Where are the dinosaurs today? They're dead. In biological terms we shouldn't discount extinction. In talking about extinction, I can't help thinking IPv4 as, originally thought, is dead. Most of the aspects of the architecture of IPv4 aren't there today. It's not what you thought it was in the same way the gigabit Ethernet has nothing to do with multiple access Ethernet. What you now call Ethernet is not and what we call IP is not. Who lives behind a NAT? Wow! Most of you do. One way or another you're sitting behind network address translators. Any application that works is going to have to work through NATs. A lot of you are now on Skype. Before that was an equally useful free piece of software called Speak Freely. The issue about it was it couldn't cope with NATs so it died. No one uses it. It doesn't work where most of you are. Why is Skype so seductively cool? It works behind NATs. All of a sudden too, the ISP industry is going into a complexity spin so now we have efforts to impose overlays, virtual circuits, active middle ware, they seem to be making life more difficult, not easier. They seem to be making life more expensive. The original value model of IP - fast, simple, cheap - is disappearing. It is none of those things and it is getting worse. So IPv4 has a problem and even on the address side it could be argued despite all the numbers and progressions, they're actually largely irrelevant because we ran out of IPv4 addresses when the first NAT was deployed seriously because you couldn't find a reasonable economic region to deploy real addresses. So today, 1/32 with your DSL for upwards of $15 a month in some places in the world, as much as the DSL access itself - now the address is costing as much as the access, an amazing proposition. What will happen when we run out of the unallocated addresses? It doesn't mean you won't be able to get addresses, it means the market for acquiring them will change so the pricing and distribution function will change. They'll always be there but the way they're distributed in the industry will change. Today we have made a very complicated world. Maybe that is mainstream industry and a natural reaction but it is very complicated. If you're an applications designer and there are actually very few of them in this room. They're in different conferences. They're over there with Tim O’Riley doing XML, blogging, BDSL or WSDL. They talk a different language and they're very confused people because those poor folk have to write applications that have to negotiate through an incredible craft of middle ware. The plumbing is amazingly convoluted. We have NATs, firewalls - you can't assume between you and me are one or two NATs, there's any arbitrary number. The NAT in front of me might do it one way and the NAT in front of you might do it another way. Getting a packet between me and you is a challenge for the application engineers. What do we do? We try to simplify the world. This is not a peer to peer network anymore. It is comment-initiated, all client server. Your IP address is not who you are. Applications are reinventing identity. Look at Skype. What's your Skype number? Your IP address? I don't think so. They have reinvented a whole new identity realm just for Skype because they had to. That's a complexity. The other thing about Skype, we're going to see it with a bunch of protocols up the top including bit torrent is because you don't have end to end, because I can't send you a packet without knowing what your current NAT translated identity is, we're putting in the middle a bunch of multiparty rendezvous agents, interceptors and redirectors. Now the NAT stay is shared between multiple independent networks. That's hard and hard is a key word for complicated and complicated means expensive. All of a sudden the promise of cheap is becoming the promise of really expense. The promise of light-weight applications that just work are becoming horrendous designs in pre-guessing what kind of strange behaviour today’s NATs are actually doing. Is this what we wanted? Should we move on? If the real promise of the Internet was cheap, if the real promise of the Internet was this is a better way of doing it because it reduces the cost of the point of everyone can afford it, yeah we should move on. If you ask any individual player, whether it is an application designer, a carrier - they'll probably agree with you in abstract but the reasons will be different. We want networks with more directly addressed end points say the application engineers. We want to reduce the per address costs. In IPv6 in theory we should be able to drop the cost astronomically. We should be able to make simpler applications that are cheaper to build, deploy and field. We should be able to allow direct peer to peer networking. Today in the ISP industry, between 50% and 80% of your traffic is peer to peer, very complicated because of NATS and middle ware. We should be able to get down to devices and make the entire industry cheaper. So yeah we should move on. But no one can actually feel the pain properly. Can't frankly, ISPs don't write applications so the cost of deploying a NAT for an ISP is actually slight so the easiest way to do it is to deploy it because the ISP's don't pay for the APs. Existing players who say, "I have bought the Internet and built the infrastructure, why should I spend more?" Say, “OK, it is a good idea to move on, just not today.” The new players - the folk just coming in, the rest of the world - speaks IPv4. If you run an IPv6-only network today, you're talking to yourself - incident business disaster. Frankly, there is no benefit in early adoption. Most of the players aren't looking 10 years ahead. They're looking five months ahead if you're lucky. If you look five months ahead you can't see IPv6. There's no early award. For some reason the pressure to change doesn't exist. The industry response is IPv6 is cool and we need to gee there but not today. Later is what you'll find. When does later become now? At what point does this final deferral smack you in the face and go, “Whoops, time to think.”? I have no idea. I'm not sure anyone does. The industry isn't able to answer the question. We know it is a good idea and know we should go there and know we should do this. When is becoming problematic. Maybe we should look around a bit wider. Maybe it is not IPv6. Maybe it is. Is there another way to do communications that offers us more immediate impetus to make this change? Let's think about that for a second. We aren't actually saying what it might be, the first question is how long would it take to think about a new way of doing communication, to doing multiplexing, to actually do the public communication network differently? We have gone through frequency division multiplexing, time division multiplexing, packet division multiplexing. How long would it take for researchers to come up with fantastic new ideas? Then you ask the next question. If you think about data communications, how much of this space is already constrained? Even if you did ask an amazingly large group of people to do the research, would the answer they come up with be the same answer you have now with slightly different design trade-offs but working in the same space? Think about this for a second. We want to switch packets. Why? Well, switching packets is certainly weak in terms of control theory. Circuits and time division switching is easier to operate say the operators, we have more resource control say the operators. We are able to offer tight performance outcomes say the operators. If you buy a 64 k time switch circuit from A to B, it doesn't run at 48 k or 100 k, it runs at 64 k. Very tight outcome control. However, the other thing about this is networks based on circuits aggregate revenue. Networks based on packets push value off the base of the network. Operators say packet switching is bad because we get less money. They're right. Packet switching is cheaper and more efficient. By the way, it is cheaper. The service models are much less constraining. There is no doubt the voice network was a one-trick pony built around the dynamics of the human voice, the loud and soft frequency rate of the voice. 8,000 cycles per second is about human voice engineering. Packet networks can do anything and they're cheaper. There's more innovation at the end and by the way it is cheaper. The whole thing is cheap, cheap, cheap. If you do packets, you'll do them because they're cheaper. What about service control? There's no doubt we're overrun with hostility. The plot about trying to understand what are the good and evil packets isn't working well. I would say no matter what you did you can't fix that problem in the network. The network can't tell the evil packets or change human behaviour or cure spam or cure attacks. Frankly, if even if you go in strong authentication of origin, it still doesn't work. Networks are not private. They're public. They're not closed trust domains. Quite frankly, any approach in networking is going to have exactly the same problems you have today. It is not a networking solution. So I don't think you can get better control capacity no matter what you did. Would routing change? It is almost like saying would number theory change? No, it won't. Routing is an astonishingly constrained approach and there's very few ways of doing routing. We understand we have to do abstraction. We don't want to know the street map of one city when we're in another city. We want to know the trunk routes and summarise, and there's few ways of doing this other than addressing hierarchies for forwarding. You can route politics or packets but you can't route both says John Klensin. I agree. Routing has its own set of constraints. V6 is the same as v4 and I suspect any approach in packet switching comes in the same kind of way. Is there anything else around? No. How long would a design effort take? V6 has taken at least 12 years. To go radically different, think 20 or 30. It would be a decade or longer. What would be the result? What you've got today. Like how many wheels does a car have? Four is a reasonable compromise, two you tend to tip over, three is unstable and five is one too many. There is no new protocol around. Stop thinking about it. Extinction isn't likely. What about evolution? The case for v6, why would I try to dissuade an ISP if it is in their interest? You're paying for and deploying NATs and getting your customers to pay for and deploy NATs, you're using dynamic address allocations and stopping customers from setting up permanent servers because they can't be addressed. The entire network and application design of the industry is based on scarcity of addresses which is a lousy place to be. We can do it cheaper if you don't have to do scarcity deployment. If the ultimate message of the Internet is cheap, you're spending too much money in v4. IPv6 is a very conservative technology. Quite frankly, there's little difference in IPv4 to IPv6. It is well-tested. It's gonna break - I doubt it. It has achieved some measure of deployment, it’s been tested all over the place. All your Windows stacks and unix stacks have been tested. If you think it's not ready yet, the case for v6 says, “No, in actual fact we have been good about this and think it should be a when, not an if question. “ V6 is ready. But the one thing I think about many technologies is it's very hard to know when to stop. If you keep meeting three times a year and on the agenda is, let's talk about v6, you will. When you talk about it, you'll think of things you should change, tweak, alter, improve, make perfect. There's certainly a rich set in v6 where we're still sending signals to the industry we're not ready yet. Bad move. If you think about it, what is the flow label for? Currently I think in the IRTF, I'm aware of four competing efforts to make the flow level used. If I'm a router and burning functionality into hardware, how should I interpret the flow level field? Maybe I should wait for another year to see what happens in standards bodies. There's others that oppose the issue maybe v6 isn't ready. Maybe we're arguing over bits. It is not quite the clear story we would like. Let's go beyond technology into business. Could we do v6 by regulation, by government in each country? It has never worked in the past and is probably not going to work here. What about the users at the edge? They've already bought their v4 device at the end, their DSL modem, the thing that speaks v4, speaks DSL, they've paid $40 or $100. Can I get them to pay the money again to throw out a perfectly functional device and get one that does v6? What's the value? Same email, same XML, same bit torrent and this and that. Why would they do it? They won't. Users won't pay more because for them it is still email and still the web. There's no new value there. The other thing is too we've actually created an industry of scarcity. There are all these NAT vendors who look at v6 and go where's my business angle? How can I sell NATs in v6? We have created a supply industry that's worth 10s of millions of dollars based on v4 scarcity. Those folks don't want you. They don't want v6. They're doing well. The network provider is spending money putting complexity in, charging high for this. All of is sudden v6 doesn't offer them the opportunity for greater revenue but for less revenue. This is not a good business tactic. Reducing your income generally means bad things so from that point of view it's not working. The industry itself says, “Over the last five years we've spent billions of dollars throwing out time switches and existing data work and we're terribly sorry and we've spent millions of dollars in IP and aren't we good?” And you say, “Sorry, not enough. Spend more.” “What? I have to buy v6 now? I'm sorry, excuse me? Why?” There's a real reluctance on the industry to invest now. There is no free money flying around. The investors want returns. The business case doesn't work. There's no incremental revenue flying around. Maybe we oversold too early. Maybe we went into IPv6 from a marketing side and now it's tired and not wired, to quote the ‘Wired’ magazine. It is yesterday. Maybe the message was too hard too early. The short-term interests don't work now. It is a long-term proposition and everything over there is working well. Meet the enemy of the good. NATs are seductive. Who cares how complicated the applications are. I don't write them, I don't pay for them. I don't care. You can plug in a NAT and the rest of the world can do what they like. What a seductive deployment model that is and it's worked. There's no doubt the real Internet is three or four times the size of the Internet you can see so NATs are working for the industry. Perfection is a problem here, trying to constantly polish the bits of IPv6 sending out messages saying we're not ready yet. The search for complexity is starting in IPv6 as well. We're trying in v6 to solve my problem not on an application level or another level but in v6 and hobby horses are coming out saying let's make the protocol more complicated. Complexity equals cost. Maybe this is the enemy and the problem here. Now the users are somewhere else. They've gone. They don't see IP anymore. The entire Internet service portfolio has collapsed into client server, hasn't it? Everything over XML, over HTTP, over something or other. Who cares what it is. How much of the network is bit torrent? How much of the users blog these days? What is going on in user land is different. If I'm client server, I can largely get around this and they do. It is independent of v4 or v6. Users are way, way removed from this and don't see incremental value in the transition so where's the leverage here? Maybe it is just business because we're a conservative industry and maybe zee to understand evolution has a problem because the business case doesn't support it. There's not enough linkage between the cost and complexity of the application and v6 in the infrastructure. We can’t see the benefit closely enough to make the investment. So I think evolution's dead. I don't think it would work. I don't think when you expect a comfortable - yes, let's change a v chromosome here and do it over a few years. Forget it. I don't think the business leaders are going to work in your favour. Let's talk about the revolutionary model, the two protocols are not the same. Why did v4 take over by storm? Was it better? Yes, but it was cheaper. I use all the wire and my operational costs were a lot lower when I originally employed v4 and it was simple and users paid for end devices and users paid for the handset and everything. It was revolutionary and changed the way the world looked and realistically what happened is it worked behind the silicon industry. It was really the phone system for PCs. That was the evolution. A really cheap phone system for PCs. That was v4. V6, I think has to look beyond PCs. The only reason you have 128 bits of addresses - which by the way which we found out from Wikipedia, a wonderful resource - was if every v6 source was equal to a grain of sand, the entire 128-bits would allow you to make three planets the size of the earth. That's a lot of addresses. This is not a PC-based technology. V6 has nothing to do with PCs. You wanted that bigger address space because you're following the silicon industry and those guys don't just make PCs. God no. Where can you find silicon today? Anywhere and everywhere. If you think about the radio frequency ID tags, they're even inside foodstuffs. The silicon industry is going far, far further than PCs. If v6 has a prayer, it is going to take the silicon industry and follow it. You have to think about the transition from a PC to things like iPods and go further. There is now in Japan a camera that's y-fi. I like the idea of an i-pot. An Internet kettle. A kettle with a piece of IP stack ware in there. Why? There is a really compelling use which I think underlies the story of why devices are important. We're in an ageing global population. Health care and everything else and population pressures have forced our population to get older and older. There is a greying problem out there. The baby boomers are getting old and we'll never have enough nursing homes. They're going to be living by themselves. How do you know your mum or dad is OK? How do you know they're alive? Every day they get up and make a cup of tea and the pot sends a message, "Hi, I'm alive." The day it doesn't, maybe you should give them a call. Maybe they have not missed the alert button by not making a cup of tea. How much is it worth? A couple of bucks. Not much. But it is a neat idea. If I'm going to do the device-based I have to think low value but ubiquitous. I have to think what the value is about. Frankly, the existing service platform won't change. If you think about v6, I think you have to look at something entirely different and look at a shift that the silicon industry understood years ago and for some reason our industry is not listening. Completely deaf. The silicon industry shifted from value to volume years ago. They don't care what happens to the chip. They just make chips. They make millions of chips every year. That's their business. Value not volume. What will drive v6 is the same shift from value to volume. Stop looking for the high value bits. They're not there in v6. Think about how you can make networks cheaper to embrace trillions of chattering devices you and I don't even know are happening. Low value silicon chatter. The revolution v6 offers the industry and the only leverage v6 offers the industry is not a better v4 network, it is a completely different network. It is a network that does volume over value and looks at how can I get the cost of packet switching - not just 10 times cheaper, probably not even 100 times cheaper, how can I get the cost of moving packets about 1,000 times cheaper? What do I need to make v6 work? Actually with v4, not only did the PC help but wave division multiplexing is fibre helped. All of a sudden transmission could carry more cheeps. It got cheaper. We need denser WDM, stuff more bits into a wire. We need simpler switching systems. I don't think using silicons to switch photons scales. We have to think about inside the networks how do we switch photons? It is going to be cheaper if we can make it work. How do we make a provider industry that got hooked on 85% on their investment understand that their future lies on 10% return or lower? How do we make them understand they're a utility, not a gold-plated piece of extravagance? Those are the leaders for v6. Frankly, the choices are extinction, evolution or revolution, I say the v6 future is revolutionary. If is going to compete with v4. The primary leverage here is not better. It is cheaper, it is bigger. Orders of magnitude bigger, orders of magnitude better. That's where we're going to be with v6. What we have to understand is you and I and the plumbing industry have more in common than we thought. We're a commodity and if we do it cheap enough, we will do a world of service. V6 is about volume and not necessarily about value. Thank you. PAUL WILSON: Thank you, Geoff. Questions? We have five minutes or so. Randy. RANDY BUSH: Randy Bush IIJ. I am a scientist at a large and early deployer of IPv6. ‘Large‘ must be a relative term because there is no actual large deployment of IPv6 anywhere. Geoff, when you talk about the billions of IPv6 addresses, you forget that 264of the address space is thrown away. What you can have is 264 LANs and we all know layer 2 technology does not scale well. Actual IPv6 address space is much, much smaller than the discussions. NATs won for a simple reason. They solved the problem. NATs are horrible. I have spoken to this audience about how horrible they are but they solved the user's problem at negligible costs. The market spoke. End of story. They already are doing NATs in IPv6. Peer to peer traffic is a actually, as you say, significantly larger. It is the number of client server connections are larger. One is the number of sessions, the other is volume, OK. And I think we may find in the long run that peer to peer volume and sessions will be high if the record and movie industry lawyers do not turn the Internet into another television. And on the matter of margin and scalability, it is the complexity that killed the telephone industry and makes it so that they only survive when it's a state monopoly or they can buy each other up so quickly in the United States and force monopoly by owning the copper in the last mile and it is the complexity which drove the margin down and it is the complexity that unconsciously the users are avoiding by installing v4 now instead of v6 because they can get a box for $25 that solves the problem instead of having to spend people time and applications and so on and so forth trying to convert to something that they don't perceive solves their problem. GEOFF HUSTON: I think we largely agree the evolutionary path from v4 to v6 isn't working. NATs are nice and solve problems. Yes, the address plan in IPv6 probably can't encompass the true vision of what I'm talking about in a revolutionary context. If you come to the policy SIG on Thursday, you'll see a proposal they're trying to claim back part of the address space that was lost, a premature plan I think was hastily put together. I think we're agreeing here you can't see any natural levers in the existing network to compel people to v6. It is not evolution I think is what we're both saying. If v6 is going to come through, it is going to come through because of a different initial client base and I'm looking at a silicon industry that says we want to talk but we don't want to talk through NATs. My washing machine, fridge, iPod is trying to do things differently. If there was any client at all for v6, I have talked about the chattering trillions of devices and say that's probably where we need to be but to get there we need to be really, really, really cheap. RANDY BUSH: Why do we want to get there? What's the deal? Are we trying to sell v6 or solve the user's problem? The user thinks they solve the problem with a stupid $25 box which makes our lives complex but then again that's why we're paying us. GEOFF HUSTON: Devices want to chatter. Why do we put a piece of silicon on every ration pack? Why is wall mart trying to kick out check-out operators and put these devices on their customer sales chains? Silicon wants to talk. It is not really a user proposition but down one level below that. The impetus silicon wants to talk is where v6 has a potential role. RANDY BUSH: Silicon doesn't want to talk globally and it is a major security problem when it does. PAUL WILSON: Let's move on to the next question. EDWARD LEWIS: I think when Geoff mentioned high server models, today is client server mostly. V4 NATs are perfect for that. Peer to peer networking obviously is going to be a problem for NATs. That's what we want to solve here. Instead of what Randy mentioned the motion picture industry and there's a lot of heat on file share in the areas, I think this is as much a technology question but will the business model of wanting to use the Internet, what occurs in uses that makes access come along, stripping client server to peer to peer. It's not a technology problem. The business has to use the Internet in another way to help push the technology to change. I think that's come hand in hand - there's no investment for no reason. Things have to be used for a certain purpose. That was the main comment. I have another sub-comment. You're talking about applications and their confusion over NATs and so on. When it first came out, I heard about scope address. Some has gone away but I think this is going against the ubiquitous addressing. SURESH RAMASUBRAMANIAN: Back to what Randy pointed out, talking about end-to-end security on v6, everything is completely secure and you don't have to worry about security but what is proposed here is largely connecting a bunch of devices that are not designed with security implications in mind because they are not going to talk to anybody. Right now I have seen cases are electron microscopes and printers get hacked into. Now when you've got this topic of convergence, you've got devices like for example cell phones getting viruses, it is a big problem. Now when every single fridge and toaster is likely to get a virus because it gets connectivity when it is intended to pop toast up - I don't know. We have to deal with the issues before we start giving connectivity to stuff that was not meant for connectivity in the first place. PAUL WILSON Let's give it a few more minutes. It is morning tea time but I'll take the two questions now and then we'll break. RAUL ECHEBERRIA: In your slides, I think that in slide number 10 you say that probably the nature of the function and price of function will change with v4. Could you refer to this? GEOFF HUSTON Sorry - the natural solution for? RAUL ECHEBERRIA: You say we are running out of unallocated addresses. You ask yourself if that doesn't mean the addresses would be - would not be available. You said yes but the nature will change. GEOFF HUSTON You're asking me to speculate what the change might be. In any other situation where you initially had a supply chain based on feeding from an unused pool into the deployment environment, when the pool exhausted, all the players start trading. Buying and selling happens. There are an awful lot of I P addresses that aren't being used or could be used more efficiently. Someone places a greater value on the address in the header. Like good humans we are and good economic players we are, the natural tendency when the initial grip feed turns off is to start swapping, changing, trading, buying and selling. That will happen. It will happen irrespective of I think anyone's policy fear. It will just happen. If we look further, what can we do about it? I suppose we can make sure the trading is legitimate and the folk who want to sell have clear titles and the folk who want to buy, buy that clear title in the same way the land titles office legitimises exchanges. Maybe that's the way the industry is heading once the unallocated industry dries up. When I said the price in the distribution function would change, we would move from a downward supply of unallocated into trading. QIAN HUALIN: I think we have the trouble because many organisations in China, they think using the private address is more safe. They think they should use the private address and each province, they also use their own private IP address. They assign their IP address very freely by themselves. The IP address may be coming from each other but when they communicate with each other they have to go to the centre. It is tree-like. Each province goes to the centre and then to another province. All three networks are private IP address. This kind of net is not available for private IP address to private IP address and other province private IP address. That's a bit difficult. Geoff and Randy Bush are experts so I hope if you have time you can explain how to deal with this kind of mess. Thank you. PAUL WILSON Quick answers, please. GEOFF HUSTON Right, the 2-second answer. I tried it to say two things and one was about NAT is and the other was application level gateways. What we're finding in folks using subtly different forms of IP and the 3 GPP environment is one, you don't translate the header of the packet and stuff in a new address, you limit the applications fly across the boundary and stop the application and restart it so yes, you can in application level gateways have the same address on either side and it will work as long as there is another identity plant at the application level that transcends all of this. It is a shocking solution. It is really ugly. It works and it is achievable. What do you do? Behind everything there is a design trade-off. We're finding working application gateways to try to mitigate the problem and colliding NAT worlds. Whenever there is a problem, technologists give you a solution. You can deploy it or choose not to. Technologies don't give you a value attitude. They're not saying good or bad. They're saying to go and do it. PAUL WILSON That's two seconds. Thank you to Thuy Nguyen, Tom Vest and Geoff Huston. Let's break for morning tea now. I won't get too much in the way of that break. I'll stress the onsite notice board is available. It is illustrated on the screen and you can find it on the APNIC home page. That's where the information is about what’s going on today and next come back at 11:00 and we'll have some coffee and so forth in the meantime. Thanks very much.