Wednesday

Note: Please be advised that this an edited version of the real-time captioning that was used during the RIPE 56 Meeting. In some cases it may be incomplete or inaccurate due to inaudible passages or transcription errors. It is posted as an aid to understanding the proceedings at the session, but it should not be treated as an authoritative record.

RIPE 56 Opening Plenary, Session 3 9:00am, Wednesday, 7 May 2008

CHAIR: Good morning welcome back to the plenary [unclear] sessions. Today we have v6-focused session. We will have a few presentations and then in about one hour from now, we will proceed with the switching of IPv4 note working and you will be on a v6 only network with no alternative unless you have real urge enter needs, in which case you will have to go outside and look for the missing IPv4.

So to start with, we have Geoff.

GEOFF HUSTON: Good morning all, my name is Geoff Huston, I work in APNIC in the research department. I am a member of the chief scientist cysts' union along with Daniel. Our job, I suppose, is to actually try and understand various parts of this network and how it interacts with our role as regional Internet registries. So, allot of this work is actually trying to sort of look at the network and measure it in various ways and under ways its dynamics, the interaction between players and policy and outcome. What actually happens when you make a certain policy? Do folk out there abide by what you intended, are the outcomes what you planned or are they entirely surprising. So this morning what I want to look at and talk about is measuring IPv6 deployment, and this is work I have done with George Michaelson and like I said, we are in the research and development department, the mad scientists brigade. So the first question is, you know, is anyone from another planet here today? Good. So you guys haven't heard, we are actually running low on IPv4 addresses, you know the rest of you can go to sleep. And this kind of slide now needs no introduction I would have thought. At some point, frighteningly close, IANA is going to run out and it's about three years from now. The way the mathematical model works is quite precise; on the 18th of January 20011 at about 11 in the morning but it is just a model, and quite frankly, the variation is actually about plus or minus 18 months. Because most of the address space is actually allocated to a relatively small number massive deployments, entire cities, entire countries. So that it's not just the next 10,000 allocations or so that is going to drain this out; it's actually about the next two or three hundred. And modelling the action of a large number of people is really easy. Modelling the actions of a very small number of folk have higher rates. We can't tell. So what I can say is, if life continues as normal, no hoarding, no nothing, you know, nothing abnormal, then the model will kind of run to about that date, but I really don't know, and it could be a fair deal closer, it could be a bit longer. But that is the day the train will run off the rails.

So you know, this is not news to you guys, apart from those three who come from another planet. Listen up. This is not news, because we knew about this years ago, we knew about this in about 1990, and if you had actually gone to the ITF in whatever it was month of 1991 you would have seen the routing and aggressing group forming, talking about the fact that we have problems in routing and we have problems in addressing, we knew this was going to happen, even then. And this is the kind of thing we devised. We needed a new protocol because we actually needed more addresses. And the idea was that as the network grew and as the v4 address pool depleted, which it would, this new protocol, v6, would gradually get deployed, and if all went according to plan, because you folk are rational people, you are professionals, you understand this, you will work ahead of the disaster and because you are such clever professionals, you would have deployed v6 long before we get to that magic point where the last address needed to be allocated. You were meant to have never used the last address.

So, about eight years ago, you didn't do what we thought. Whoops. Because what you were meant to do was start picking up on v6 even then and gradually, without much fuss, without involving users in panic, without panicking the rest of us, you were meant to have quietly got on with your job because you were meant to be the professionals. Unfortunately, you didn't. Bit of a stuff up, because we weren't meant to be here today talking about this. V6 was not meant to be a grand experiment. All of your machines were meant to have been running v6 a couple of years ago, remember that? So, you know, this was never meant to have been. You were meant to have completed this transition a little while ago, and the trains were never meant to collide this way. So here we are. V6 has as much deployment today as apple talks seems to be the popular Word out there. Even DECNET phase five has more deployment. Something has gone desperately wrong. Because as far as we can see, the v6 deployment metre is down there at the bottom. This is now a real problem, because whatever transition is going to have to happen, happens in that grey space where you have got no more v4. So if you are saying to yourself, as we continue to grow, we will dual stack, and we will run v4 and v6 simultaneously, then most of the people in the room today who have v4 addresses are just fine, but the Internet is still growing, isn't it? It must be, because we are consuming 16 /8s every year right now. We are consuming hundreds of millions of addresses. So the people who are joining the network in the next year are as big as this room again. But we have got no v4 addresses to give them so what does dual stack mean for them? How are you going to talk to them? What this they are the next Google how are you going to reach them. This revised plan is looking just a little bit sick. So, we need to revise the plan, that is pretty obvious, and we need to strap the jets on the plane because something has got to work.

So the real question here is not that I have a plan and you should all implement it, because I don't. And I don't think you do, either. So the best I can do is to try and understand precisely where we are today and try and understand just how much v6 is really out there and what kind of work we have to do. So, we actually only have a couple of years before life gets very, very tricky. Because we can do dual stack until around 2011 but beyond that, the v4 part of the dual stack of the new deployments is going to get just a wee bit tricky. So let's try and understand how much of the Internet is capable today of running v6, how much work we have to do. So you know, what is the revised plan? How much v6 is actually being used today? Now, being network engineers, the interest digital approach is to instrument your network and count packets, yeah? So you sit there and count the number of packets with protocol v4 and v6 and if you do this with MIBS and the right piece of MSP and if your router vendor actually supported it it instead of saying so, if you had all that then you could actually measure the way v6 is going, if you believe the MIBS. But unfortunately in APNIC or maybe fortunately, nobody trusts us running a big network. We don't. We run a tiny network. We don't actually have access to a big production network because people believe that data is confidence, fair enough. Try longterm, not just a snapshot, but say four or five years' worth of data of the amount of v6 packets every second for the last four or five years, I actually couldn't find any in my quick glance through Google or it couldn't find any. Assuming that someone is collecting this data, and I am not sure they are, we didn't get it, we haven't got it. So we have got to guess. Well, maybe we don't have to guess. So we bend the question a little bit, because APNIC and RIPE and a few others we do actually collect a lot of data. It's just not packets. We actually collect BGP table data every single update gets recorded. Look at the risk work. It's a lot of information of the we actually record every single query to our DNS route servers and we have web servers that have is a reasonable amount of traffic and some years ago, because we are professionals, we actually decide today dual stack this stuff, so have actually got v6 and v4 running. So the real question: Can we interpret the data we already have to figure out how much v6 is out there?

So, what does the data we have tell us about an option. So you know, we can look at the v6 routing table and Jean camp did this yesterday. This is a funny table because the six bone makes it noisy. If I had the magic laser pointer you would see that break in 2006 was actually the first break was the ghost routes disappearing and the second break right in the middle of the year was the v6 bone networks sliding away. But, oddly enough, noise notwithstanding, 2007 and 2008 looks a little bit funky. You know, 300 people heard the message. That is a lot of people. 300 people started advertising v6 prefixes, you know, the slope up there is indeed accelerating. It's just not a lot. But, encouraging signs. What about v4? Well you know, same old Internet graph, starts at the bottom left, goes to the top right, bit of noise in the middle. Look at the scale. So, since 2007, we added 67 thousand routes, not 300. So let's compare the two. Look familiar? The red line at the bottom, guys. So what does this tell us about v6 deployment? Not a lot. Like I aid, apple talk, tech net phase five. Let's be nice and charitable and do the ratio. Let's not just talk about it in [unclear] but measure the amount of v6 in the routing table compared to v4 in the routing table size and now that optimism is actually a little bit weird because there was relatively more v6 routing table entries back in 2004 than there is today. The 6 bone was more successful than the real thing, relatively. This is a bit bizarre, it's not expected. But the good news, the good news is that since July 2007, the v6 network is growing faster than v4 in BGP table size terms. It's grown from approximately 3.7 percent to approximately 4. Sorry,�0.37 percent to 0.45 percent. So you know, three parts in 1,000 to four-and-a-half parts in 1,000, good news. But what does that mean? This doesn't work for me. And the reason why it doesn't work for me is that for each originating AS and v4 you see eight or nine announcements, the v4 routing table is incredibly fragmented, the v6 has approximately 1.2 announcement per originating address. So when I try and compare the two I am not actually measuring v6 deployment and the figures don't mean anything; I am actually measuring the relative amount of damage you guys do to routing, and you guys are professionals, you stuff up that v4 table like nothing we ever dreamt of. You can take a /20 and produce /24s in the blink of an eye. You can take a /8 and produce /24s in the blink of an eye. So all I am measuring is actually fragmentation and the relative levels. Not very helpful. So what I can say, that is me, I heard that comment what I can say is that it seems that there has been some increased interest in v6 in 2007 and that is good. But I really can't quantify it. I can't get 25 percent inside this number, it doesn't mean anything to me or you. So, that is OK.

Let's stop that one and look slightly else slightly different direction and rephrase the question to capability. If, as is happening to you in, what, 20 minutes, v4 turns off, you know the great v4 media strikes the earth and that is the end of v4, who is the mammals out there? How much of the Internet is going to survive and still breathe air? How much of the Internet is capable of running v6? And one of the ways I thought I might answer is actually look at autonomous system numbers because these are closely aligned to companies and operators. How many of the service providers out there in v6 have v6 routing table entries, same kind of curve, bottom left, top right, this is v6? In 2007 we had 650 ASes and today we have got 930. Great. Better curve actually, because the 6bone stuff doesn't really figure any more. You see that slight correction in 2006 but I think you actually see a more consistent picture of players and the uptake of v6 is steady. All this work is actually giving you some outcomes and about since mid-2007, more people have been picking up attention. It's looking good. There is the v4 graph. Since 2004, it's just phenomenal. V4 steams on like crazy so every year there are at least, what, close to 3,000 new players. Every year, there is double the number of players in v6. So v4 is still screaming along right now. If there is a cliff out there, you guys are going to see if the pig really will fly, because you are going towards that cliff edge so fast it's super sonic, the bastard might just do it. This is speed. But relatives are actually again looking OK because if you take the ratio, the number of ASes joining v6, small though it is, is faster by comparison. We have gone, this is.3 percent, this is 3 percent (isn't) since 2007 we have gone from just below 3 percent to right now, at three-and-a-half percent, almost. That is a lot. That is lot. That is one in 30. So of this room, you guys are good, what is happening with you? And that is not a bad number. It's not just the first two people in the left. So, interesting. Number of ASes risen from 2.5 to 3.3, I am encouraged, you should be, too. But this is weird. It's actually better news than we thought and worse news than we thought because although there are 28,000 ASes in the routing table today, only 4,000 of them, approximately, deliver service to others. Only 4,000 are in the middle of an AS path, only 4,000 of them have a number of customers and a number of up streams. The other 24,000 are stubs. They appear to be very small folk who are multihoming right at the edge. So the real question is, if we divide these into two pools how many of the transit folk, of the ISPs, of the professionals have picked up the message? And what about the end users? Now, this is interesting because 13.8 percent, 14 percent, you know, bigger number, are actually picking up this message that they are running v. 2346 some way. 13.8 percent of the transit ASes actually original v6 so some picked have picked up only 175 percent of the low end of the AS market has picked it up. Good news /bad news. So, does this mean that 14 percent of the Internet could run v6 if we needed to? No. No. As we heard yesterday from [unclear], a lot of these folk are doing it in their lab, they are announcing but it's not an S it's not that their entire network is capable. We actually don't know, but what they can say is that of the folk who are playing in v4, of the transit folk, around 14 percent are also playing in v6 to some extent or another. So OK, the capability measure is encouraging, this is cool, I am encouraged, you should be, [unclear]. But I am still not happy, and neither should Detlef be happy. It still doesn't give us this idea of when do we really know we are doing v6? How much is it being used?

So, you know, APNIC also collects other data and we run reverse DNS servers. I have never understood that and I am not sure anyone does. I don't know which applications demand the fact that you translate the number back to a name but nevertheless, a lot of you do, for reasons I don't know and understand why. Because you do we are logging who you are. We are logging your source address and what you ask for. So we have data that goes back at least four years, that actually listens to and logs what address you are trying to resolve. V6 or v4? Interesting. And we also have divided up our servers, we have a few of them around the place but one set are predominantly Asia Pacific and other seconds for /8 which is predominantly in Europe. So that is a neat place to actually start looking. Let's look at the query load every day, and let's divide the queries up into queries about v4 and about v6. Now, I am going to be kind; it's a log scale. So you can actually see down the bottom of the v6, otherwise it would be the same graph we saw before, line down the zero, where is the data? This is a log scale. So interestingly, there is a bit of a difference between Asia Pacific and the rest of the world in the v6 queries at the bottom. Almost one order of magnitude.

Does this mean that we are a very advanced bunch of humanoids out there, if you really want the professionals, you are in the wrong part of the world here? I don't know, I don't think so. I actually think that the way DNS works, the secondaries is is a long way away don't get asked as much and the generally, the old 80 /20 rule in telephony that a lot of the traffic is local is being mirrored here as well and we are kind of looking at localised traffic. So that is not unexpected that we see that. So interestingly, even though the network grew like crazy the query load hasn't really grown like crazy even in v4 or v6. So maybe machines are getting smarter [unclear]. Let's now look at the ratios. So this is the same kind of thing, 0, 1, percent, 2 percent, the ratio of v4 to v6 queries. And there are these two gigantic anomalies that just stuff up my scale. I am suspicious. The data is so sensitive to someone doing an experiment over a couple of months that I am not sure I am seeing the effect of millions. I am not actually not sure how to interpret this data.

What it says is that around two paths per thousand is v6. And not unexpectedly, the query load in Asia Pacific is higher for v6 than the rest of world in Asia pass I can and the queries load has increased a bit but I am not sure what I am measuring. Because v6 and the DNS and v4 and the DNS are weird. There is forwarders, there is caches, I am not sure if the TTLs are being treated the same way and quite frankly the more you bang at something the behaviour in DNS is different so am I actually measuring the relative behaviour of DNS, of heavily queried things and lightly queried things? So, I look at this data and I go, it's just a data point, it's not a big indicator. I am still not at the 25 percent, I am sorry, don't go looking in the DNS. So I am getting a bit desperate so I go to the last thing that I think is actually real, that I have got long-term data to. Unfortunately Google won't give me their data, they won't tell me who is querying them every second. Damn it! And they are not dual stack so I am not interested. What I do have is dual stacked apnic.net for four years every query and RIPE were kind enough to also let me have anonymised their data, I don't care who you are, I want to know injure source address I don't care what that is apart from v4 and v6. I counted the number of distinct visitors every day. So if you are a [unclear] and you pummel the site you still only one source address and if you stay there for one second or all day, they are such exciting websites, you are still only one address, so I don't care how many times that address visited. I just care about the number of unique source address that is came to the site that day. It's not hits or anything like that, it's just simply how many v6 people or v4 touched me each day. What is the ratio. So this is kind of interesting, and the question is that I think I am answering, when there is a choice, how many of the end host out there in a relative sense will use v6? Now, what we do know is something else, too, which is actually a little bit good. If you went to, on our Microsoft Windows XP and said install v6, then the browser application in particular will prefer v6 over v4. If you actually look at what it does in XP if, it's able to do it in v6 even in 6 to 4 tunneling, it will try and only when that doesn't work, 19 seconds later in the worst case, will it go to v4 so. This is actually not bad because this triggers that behaviour, the dual host side will actually trigger the ones that, if they really don't want v6, it will come up by preference, so what percentage of end systems will prefer v6 when there is a choice? So here is the ratio. Over four years, every day. That is not a bad baseline. Isn't that interesting, that from 2004 to 2006, the relative interest in v6 was slowly declining, slight but slow. The amount of interest in v6, particularly over 2005, 2006, was slowly declining. Which actually I think is right. If you remember the academic and research networks in 2002 and 2 now 3, were all doing tunnel brokers and there was a lot of activity and then it slowly died out a bit, that waive of interest in v6 did decline because the commercials out there weren't picking it up and nothing happened.

But look at 2007, it's a different pattern and look at the early part of 2008, that is getting interesting. So OK, this is just one website, whoopy do!. Here is another. So this is the RIPE website. Now, they stopped the data set they gave from early 2004 to 2008, again it's the same.2 percent. That decline, if you look very hard and squint a bit, that decline is there and then in 2007 a pick up in interest, and again, coming up late. Put the two together and actually get the opposite effect; that you guys in Europe are more professional than news Asia Pacific. Your rate of v6 is slightly higher, the brown sort of on slightly higher than the blue. What it does say which is encourage something that my APNIC data [unclear] RIPE data point agree. And if someone else is willing to run my scripts and they have a dual homed website that has been running for years that has at least 5,000 hits a day, see me after and I would really would like to see your data. Run your script and tell me the answer because I would like to understand if that is just us or it's more common. It looks consistent day by day but you probably have this data and if could you let me see the answers, I would be interested.

And what I like is the data is so sensitive that you know, you are going to get logged today because oddly enough, in APNIC, you go and you turn on v6. And you go to RIPE and you turn on v6. So those are the weeks of the APNIC and RIPE meetings, believe me. What happens when you go to the ITF? You either don't turn it on or you don't visit these websites but it's kind of interesting the sort of positive feedback, I seem to be measuring real people. So I am there. And look at the trend line. You know I have sort of the vaguely put one in and there is a bit of eyeballing rather but it's not bad. What it does say is in a since 2007, we actually are having more deployment out there. So this is good. It's still only.2 percent, it's still only two parts in 1,000 but we are getting up to four. So you know, is interest picking up again? Is this all about auto tunneling? Is advice at that having an impact because advice at that do this on by default. So OK the next thing I look at is how much of the access was native and how much was tunnel [unclear] so 100 percent, this is one to hundred, and the red is 6 to 4 and the green is Teredo. And Teredo is 0 until late 2006. So, I think I am picking up some real data. Interestingly, it hasn't grown. It's about 4 or 5 percent, but it's not growing. Is anyone buying vista or am I measuring the wrong thing? I am measuring the wrong thing. The interest thing is 6 to 4 is actually remarkably pen [unclear]. What it does also say is that at least 20 percent of the v6 is sitting round there in sites that don't have native v6 at access level. They have to do the first leap out across a tunnel. So there is still a fair deal of tunneling in v6 from this data. RIPE data, more 6 to 4, interesting, than an APNIC. And the same sort of slight Teredo. Put the two together and yes the clustering is pretty evident. 6 to 4, 20 percent of v6 is tunneled at this point. Teredo, around 5 percent. Not bad. So 25 percent of the clients are tunneling one way or another but I am not measuring vista. And this is a really weird thing. The way vista works is not the same as XP. If you have a dual stack server like RIPE or APNIC and the first thing you will do is if you have native v6 you will do a v6 connection. But if you don't have native v6, vista has a choice, try v4 or try tunnel via Teredo in XP - the tunneling approach was preferred before v4 but in vista it's the opposite. If you don't have native v6 and there is v4 native it will do v4 before Teredo. So, I am not sure I am measuring all of the capability (so I actually need to give a different kind of website today to measure the Teredo and if someone wants to help me play that game, particularly something like IPv6.Google dotcom because the one thing that will flush outer aid dough is is a v6 only server that is popular, that has a lot of hits because then you will see because I can't do the v4 option, Teredo will get invoked every time even if you haven't got v6, I am interested if others are interested to, see if we can develop some measurements. So where are we with all of this? Where does all this data leave us? Warm and fuzzy or feeling a little bit worried. There is some warm and fuzzy out there. Consumers haven't picked it up, there is no commercial demand that really is kicking in, but nevertheless, a proportion of the folk who actually do Internet servers are active. That doesn't mean they are fully kited out their network, that not at all. But what it does say is that between one and hundred people in every of those companies is busy bag v6y advantage list and has put out something on the network the core of the network probably isn't fully do you mean stacked but it is saying that service providers are listening to us. That you guys are hearing the message and saying well I can play and at least understand what the problems are right now and that number is growing and that is cool. But out there in host land, despite the efforts of folk like Microsoft to embed v6 in every single system out there, we are still only two to three parts per thousand and I think I am over measuring. Even with Teredo I think my number is high. Because of NATS. Right. Entire sort of country of DSL users, one Nat in the middle, I am measuring the Nat address. In v4 I am not capturing all of the v4 diversity. So I think NATS in v4 under count my v4 measurements. And the other thing is, too, that you know, as popular as www.ripe.net is and it's amazing popular APNIC is it's not a very popular website. Apart from you guys and I am not sure even if you guys think they are exciting, to everybody else they are not their favourite place, it's not their favourite home page. I am capturing tech weenies, I am capturing the people who are going to be the lead adopters, so I start discounting even there and I think I am actually down below one part in 1,000. So, I think that red line is accurate. I think where we are is .1 percent and I think we have only got somewhere around three years and however we are going to do this, you are going to have to dual stack with an exhausted v4 free pool, which means you are going to have to be phenomenally creative as well as the professionals you obviously are. This is a really tough challenge, and I suspect for all of us it's going to be the toughest challenge we are going to face. Because this is hard. But we do at least have an answer for Detlef; I can tell you when you are at 25 percent I think, I think this approach is actually giving us some decent data about real, real deployment and capability out there in the network. And I think we can keep on doing this and give long baseline measurements, so it [unclear] it does ever kick off up there, we will be able to know. At least that is encouraging. Thank you. Questions?

(Applause)

AUDIENCE: I am not sure, you realise that what you mention about the behaviour with vista is correct but only with Internet explorer. If you are using other browsers, they don't necessarily follow the policy table so for example, opera it's preferring actually Teredo if it's available to IPv4. It's true that changes from XP to vista, and I some other applications that modify the policy table. So after they the policy table becomes modified, Internet explorer will begin prefer Teredo. It's a rare situation but I have seen that happening. The other point is I really believe and I have been doing measurements from a long time, I really believe most of the IPv6 traffic that we have today probably about 90 percent, its transition and it's not because the web servers, it's not because services like mail servers or web servers or DNS or something like that thank; it's more because of applications like bandwidth [unclear] and so on. The other point is if it's more a question than a point, is do you have either in RIPE or APNIC actually 6 to 4 because that will probably change a lot, the behaviour, because when there are relays on your own sites you will get more hits using those transition peck [unclear], because maybe some relation other places are not allowing the clients if they are.

GEOFF HUSTON: Let's go through these one by one. Who runs Mick row soft software. Who runs exploder by default. You know most of the vast folk out there don't muck around with their browsers and we are measuring the millions here so I know Operamat does it differently and fire fox when you are trying to get mass marking deployment numbers, the shrink wrapping is the wrapping, the exploder behaviour I understand there are anomalies I think I am measuring the right thing. Your issue about 6 to 4, my there are very few policy constraints on that and although there is some advice that if you put up 192.99. no it's 88 or whatever 6 to 4 number is, you merely meant to serve the entire world and if do you a restricted thing, take it away you are not annoying people. We haven't got relays at this point. Thinking of joining up with a programme in New Zealand, that is more about fostering v6 than it is to make our servers more reachable in tunneling because there is reachable as they are reachable right now. That won't make no difference. So part of this is I have deliberately phrased the question so that the data I have answers it. You can phrase other questions so that the data you have answers it, yes. But I have phrased it in such a way as sort of is the data that I do collect, can I interpret it in a helpful way? Can I provide some indicators of the amount of utility of v6, the amount of capability based on simple observations, because I think if we are going to measure v6 deployment over the next few years, and we are, it's going to ablot easier to get simple measurements that are consistent every day, and then we will be arguing about the measurement, whereas we really should be figuring out how to do the deployment.

AUDIENCE: Bill managed and I am not organised but I know where my socks are. I have a couple of questions. Can you provide some information about your source for DECNET phase 5 deployment, because I am not entirely sure I agree with you.

GEOFF HUSTON: We had two machines back in the deep south of Australia there that were run running it. That is my source. Two whole machines

AUDIENCE: So in fact you are probably right there. The second has to do with the size of the routing table consideration. Last time I checked, most people are injecting, as a v6 route, something that looks like the equivalent of the entire IPv4 address pool.

GEOFF HUSTON: Yes, /32, yes.

AUDIENCE: Right. And people injecting something a lot smaller as a v4 entry. So the number of useful entry number of useable addresss in a v4 route announcement, tends to be a lot higher than the number of useful entries in a v6 route announcement. It would be interesting to me, because you actually got this dichotomy now saying I have got this number of routes but I have actually got these number of hits so you are looking at end nodes using IPv6 and the number of routes so I would be interested in the number of useful or used addresses per route entry. As long as you are diddling around in the data. That would be something I would.

GEOFF HUSTON: No one wants to give me their traffic data. I will take it, you know, when you give me a feed of packets out there on your networks and I will do something. I just haven't got usage data I understand what you are saying you need to run a network to get stuff of where this traffic is going

AUDIENCE: Does anyone in this room run a network?

GEOFF HUSTON: Nobody is admitting it because I am about to leave on them

AUDIENCE: Apparently you are in the wrong venue. Go to operations venue form.

RANDY BUSH: Geoff, I think from the last two questioners, I think part of your presentation either I misunderstood or they did, could you please focus back on the fact that I think what you are measuring is the capable hosts, not the amount of traffic, not the BGP prefixes, not all that crap; you are saying, if I understood it, that these are the hosts that are capable of v6. If, today have v6 on, OK. And if, tomorrow, bit torrent come out with a v6 preference version, if you were measuring traffic would you see this great big anomaly due to those hosts are enabled, so you are talking about you are trying to actually measure the deployment of v6 capability, not the utilisation of v6 traffic, not the announcement of prefixes, not the density in prefixes, et�cetera. Am I correct

GEOFF HUSTON: You are correct, although I did give you two pieces of information. The host counts against the web server is about capability of v6, yes. Not traffic volumes or anything else. Just if v6 was all you had, these hosts would actually this proportion would get there. The other thing I would say which I also thing is an interesting metric, the 14 percent of the ASes, encouraging news that they have at least picked pup the message and know how to spell 6 and that is better than.2 percent. So two messages I think is really what I want to leave you with

RANDY BUSH: I think these are good but they are much better things to measure.

GEOFF HUSTON: Thank you. Thank you very much.

(Applause)

CHAIR: Next up we have Fernando, on another way of measuring v6 deployment. We are about 20 minutes away from v4 turn off, so if you haven't configured your machine, make sure that you have a DNS server configured. There are some copies of the leaflet and also available on ROSIE, use it while you can get to it.

FERNANDO GARCIA: Hello, good morning, this presentation is made in cooperation in a corner. First of all, I would like to say you know this presentation, this is the third time we made it. I would like to know how many of you has seen the previous presentation? OK. The other are in the middle, perfect.

OK, just I will try to not to repeat what I said in all the other presentations. Because it's almost not interesting. For those who do that, they know it, we try measure the IPv6 in a different way. The question was if our wives wanted to connect to the using IPv6 what they will see and our wives by the way are lawyers that have no idea of computers, IPv6, this kind of stuff.

So, the reason is that, yes, this is our wife. We wanted to give some real measurements because we are speaking about deployment, table size all this kind of stuff, but as you have said previously, we wanted to know IPv6 is really something more than Terrical or even bigger than a [unclear]. This is the measures we actually do. First there is no rocket science here. Anybody with IPv6 connection, well, there is some rocket science there, but everybody with IPv6 connection unique tools like tell net can do that. That is by the way, how we do this. We take three things: We take DNS, we take email and we take web. As a source for the URLs, the idea is to check the more test with servers in the world and DNS servers and check their email servers. For the source of the we use Alexa. It's rank of web pages visitors. Of course, it's a ranking. And while we ask for DNS we ask for AAA entry for DNS. And we try to check the DNS with IPv6, if there is a web page with IPv6, OK, we get an AAA for the web page, we try to connect, TelNet to port 25 that is. When checking the web page, web server, this is way in theories to have the same name for IPv4 routers and IPv6 routers but we don't know why people try to we also check for IPv6.example dotcom and OK we get a lot of results and of course you can check these pages. The result of this page is calculated daily and you have measurements there of how the IPv6 in the popular and we check the route servers.

OK, this is the gross results. This has the purple line is how much IPv6 web server are related to the total web servers, in fact actually the last measure is about 5400 web servers. From those, that is also have IPv6 address, but no, sorry, 2 in the last the green area is what this number of servers that really there is no measurements because we didn't check out, we didn't have very good we didn't how many servers really. Not too much, I know.

Compared to their measurements, we discovered with we have one more site in all the countries, we check. We will call it the IETF meeting effect. We think these server appear more or less when the IETF meeting, it was going to make an IPv4 blackout. And we will see about that server later, but I think you know it. These are the big numbers. Actually, we check 109 countries. 5400 domains and the unique web servers increase, that is good. But of those increase, useful information we only get 10 and from the others, some didn't answer, some delayed test so, we have really 10 servers that have real that real information on IPv6. Not the IPv6 server this is our test servers. The same information from SMTP and DNS. SMTP actually decrease compared to previous measurements. DNS server has a big increase. These are the cumulative measurements along the time. Take care this number is units per 10,000 so we have 300 DNS servers for each for each 10,000 servers. Some fun and facts. IPv6 Microsoft is still in there DNS but they still don't have any real web server. Same happened with HP dotcom but HP is not so much it doesn't appear so many times. The Spanish newspaper he will money dough.es not only by Spain, they have DNS entry but the web server is actually done not answering. But it has been replaced by telegraph.nl so if you speak Dutch actually you are you have you are lucky.

OK. Actually, the more is light, that is a real the centre of Telco by the way: The general result is that we have increased the number of sites, real sites from total of 0.16 percent, wow, we have 0.12 percent of real sites. So some servers that have real information. And these are the servers. This one is the new server, as you will have discovered. And yes, you will discover it in three minutes. The Telegraaf newspaper. OK. Par an dotcom as we tell in the last presentation is South Korean server that is used bay lot of Muslim countries, we don't know why. I think somebody should investigate on that. OK. Rich and famous. Actually we have seen some increase in the adoption from the RIPE on ARIN, APNIC and RIPE will have mixed servers in the next meeting. And the same thing. DNS here is the thing that has more improvement along the way because we have actually eight root servers, we have IPv6 addressing, and six of them are paying their name in /domain /named root, so the B and L appear as having IPv6 address but they are not included in the name root. So actually we have a full hierarchy of IPv6 address in DNS. That is good, because as you know, DNS is the base of all the other names.

Regarding of the old TLDs, including countries and generics, we take a total take 281 some increase because some decrease with four servers and one server, but in general, you see that we have improved a little on the situation. We have still have 92 countries that doesn't have any IPv6 server, but that give us almost 200 that can be reached by IPv6. Some interesting generic TLDs. And one interesting thing recent survey about 75 percent of the European area country code registrars for country codes, affirm that they have some methods of setting records in IPv6. That will complete the hierarchy from the root TLDs to the company server on DNS. We haven't checked that. It will be interesting to discover if that is really or whether a survey.

So the most used server, DNS server in IPv6 is still sun I can. We have other ones. OK. These are simple information. RIPE doesn't have any chains and I am finishing now. What will regarding getting an all this information, what will be the Internet powered by IPv6 for my wife? In the first meeting, we said Lord Howe island in the Pacific 350 population and 700 kilometres from the nearest civilisation. But we say, no, IPv6 is very we think it's like Greenland. OK. But RIPE 55 we made a beautiful statement stating that implement IPv6 and taking into consideration how the world received this statement, actually we think that in an IPv6 nobody listen to us. OK. Thank you.

(Applause)

CHAIR: Questions?

AUDIENCE: [unclear], I am here because my wife has IPv6 only at home and I can leave my home for two or three days, it's allowed. We have v6 and v4, native v6 at home. V4 by connection to a neighbour DSL, mostly cut connection, doesn't matter. So it's possible to do it, it's possible to do it without notice. The main reason that it works is we have working infrastructure on v6 and we have a proxy on dual stack incoming and outgoing. If you do your measurement before for web servers, I can do it in our local network and I found about 60 persons 60 percent of all traffic is IPv6, because we hand out an IPv6 to each customer unless the customer complains, and we do it for web service and all the traffic is locally. This is possible. So I think you are a little bit pessimistic, it's possible to leave home at v6 only, go out and it will still work.

FERNANDO GARCIA: I think that that what you said is real, but is based that your provider give you some kind of broker mechanism, some gateway mechanism. If you don't have that mechanism, there will be not so nice.

AUDIENCE: I have seen a measurement and I can't find it, I was just trying to Google for it and Geoff suggests maybe it was Patrik F�ltstr�m that has it which actually went to the members' sites of the IPv6 forum members. And measured how many were reachable. There was one. The point is what is not capable. OK. With the junk we have out there, a smart engineer can almost get this to fly. The point is how much is actually done and how many end users are actually going to make it when things get tight in a couple of years, and remember, Geoff said 18 months variance on his January 1st 2011 estimate. That is one year from now, is the worst case. OK. And the fact is we have test as far as I can hear today, is we have some test deemployments and we have some stuff out there with people playing with it, but it's not well deployed. And Geoff's fear, as saying, what the hell are we doing, is very real.

CHAIR: Thank you. Many things are possible but I think this talk was talking about what was actually happening, just like Geoff's. So with this, please James, RIPE NCC laugh brief description of how the v6 only network is set up and then I think we are going dark. Hopefully not too dark.

JAMES ALDRIDGE: OK. This is fairly short talk on the v6 network at this meeting. A bit about the v6 hour and a few graphs.

Basically we are going to cover what we have done in v6 for RIPE meetings up until now, what has changed at this meeting, the transition mechanisms we have in place so that 99.9 percent of the Internet random number is still reachable. Quick drawing of the network, the bit of invention about the v6 hour, where in the hotel the dual stack network coverage will remain. Thanks to a number of people who have helped out with this and a bit of time for questions, as the network is being turned off.

Up until now, RIPE meetings [unclear] just had a dual stack IPv4 and IPv6 network. This is the SSIDs RIPE MTG for 0 for this meeting, we're using 2001, 4220: 1 /64 on that network.

And that is classic dual stack v4 /v6 network.

For this network we have added two additional and I put IPv6 in quotes in this one. We have RIPE MTG v6. You may have noticed there is RIPE MTA v6. That is a v6 only network, there is no v4 on that network. 2001: 4120: 1. 1 /64. We have an issue with Windows XP it, doesn't do DNS over IPv6 at all so we have provided another network which has local RFC 1918 IPv4 address space providing trance fort a local DNS server. Transition mechanisms we have a Cisco router doing NAT PT and a box providing DNS gateway.

NAT PT, just quickly, it's the network address translation and protocol translation described in RFC 2766. We are running IOS version 12 .4 T5 which was compiled on the 15 of May so it's very recent code. I am told that it's 12 .4 [unclear] T 3 or later should also work. The mapping between IPv6 and IPv4 means that all connections to v4 sites from this network will appear to come from 193.0.29.3 a single IPv4 address. I will go into more details of the configuration in the IPv6 working group [unclear]. We have to do something with DNS. We run the tot D software, it synthesizes which only have provide A records. The network looks kind of like that with all the dual stack most of the details of the dual stack stuff removed. The two v6 only networks down here on the right, the XP network, tot D, same machine connects to both providing the DNS proxy. [unclear] up here is our normal dual stacked resolver on the normal public network. Dual stack network is generally handled by a couple of Juniper 232,300s, in between the v6 only network and dual stack world and indeed the v4 only world is a Cisco 7301. The Cisco also provides the HCP for the XP LAN and version 6 for both LANS for those clients that support getting additional information like named servers through DHCP.

When the IPv6 hour [unclear] a few minutes, we will switch off dual stack wireless networks except outside the wire room at the far end of the lobby area here. That is the RIPE MTG and A networks. Wired connection to the terminal room will remain active. If you use Windows XP you should be connected to the right MTG v6 otherwise connect to RIPE MTG v6 network.

The quick plan of the hotel. The area pretty much covered by the wireless LAN shown in green there, we will be turning large chunks of it off. If you want to carry on working and you don't trust the Nat PT set up or your application doesn't work through that or for whatever other reason, make your way down to the end of the corridor now. This is your last chance to configure laptops. The configuration details for most operating systems can be found on ROSIE on or these orange sheets. Before we switch off the network, I have got to thank a few week. Philip Smith, who together who spent a lot of time with me over the last couple of days debugging IOS and Nat PT configurations. Randy Bush who pointed us on v4 only and v6 dual sites to test against. The RIPE NCC staff including but not limited to Eric, who is hovering down there ready to turn things off, and Ruben who added extra virtual machines to provide the extra services at short notice.

Any questions? OK.

RANDY BUSH: That poor little Cisco 7301 who is doing soft forwarding of all the NAT PT stuff, how is the CPU doing?

JAMES ALDRIDGE: I have got some graphs to come up. You will see some graphs N fact you can probably see some graphs now, if I can find them. My browser works on v6 fine. That is our 7 percent or so CPU load at the moment.

RANDY BUSH: Cool.

JAMES ALDRIDGE: V6 basically I am hanging on this slide until all the v4 is off. Now. We will switch the networks back on again at the end of the coffee break, so in the next session there will be normal dual stack service. If you need any have any problems, people wearing blue badges are the technical team here and we should be able to help. We have sort of helped set other people up, so if you have any questions, during the coffee break, rush outside, find one of us in blue badges.

A few graphs. I showed you the this is the traffic in and out of the v6 and the v6 XP LAN. The other graphs I have to show you at the moment are the number of clients associated with each of the SSIDs. The red is the normal dual stack network. Green is the real v6 and blue v6 XP network and I will be coming back with these graphs which are updated every minute in the at the end of this session. And quickly back to my presentation again. If Eric can wave at me when the v6 is turned off everywhere. I think I have one more slide to go. The next slide comes up when I get the nod that v4 is off. OK. I am told that everything is off. So, I have been asked to give you this URL where you can answer a few questions and have and maybe win a prize. If you are not on v6 you can't answer the questions. And that is the end of end of my presentation.

CHAIR: We will be waiting for you to come back at the end of the session.

AUDIENCE: It seems for the people who were on IPv6 already using the RIPE meeting v6s S. I [unclear] we lost external connectivity.

SPEAKER: Just now?

AUDIENCE: So something went wrong in the lab.

RANDY BUSH: That is exactly with a what I wanted to avoid and did not succeed in doing so. OK. You don't want to know how much I paid to have the network turned off so that somebody would actually make the mistake of listening to me. Excuse the slight commercial because people don't understand where I work and I actually live in Tokyo, I am here because I used to actually work in the RIPE region, etc., etc., and I still have friends that I want today eat dinner with. IIJ was originally just an initiative to get Japan on the Internet, Asia and US backbone, no commercial customer base, NTT has that market. We are Internet, we are not a tell [unclear]. We don't, you know we are heavy IPv6, we do not use MPLS etc., etc.. we were the first commercial deployer of IPv6 and the WIDE [unclear], etc., code basis which most of you are running came from partially from my Co. work coworkers.

So this is the short presentation I am going to give in one foil, you unfortunately [unclear] listen to more but this is what it's going to be. As we know it's going to run out, it's not news, frank, it's now over 15 years ago, Geoff's stuff in detail. IPv4 is going to go to a trading model. The registries will become title agents, not allocaters of IPv4 space and the tools for doing this are being developed at the registry today. As Trudy's presentation and the other presentations today indicated.

What should have happened, as Geoff says, is you know, v4 ran out slowly because we deployed v6, the cost of getting v4 relatively low, because we had v6 deployment, well we didn't. OK. So v6 actually didn't deploy, and friends of mine say that this is the most optimistic they have seen me, the fact there is an inflection point here, and so the IPv4 pool is running out very [unclear].

So if you think v6 is being deployed well you have already seen a number of presentation that is disabuse that. This is a different measurement. This is the v6 prefix count, yes we are panicked in Asia. And this is the allocations in the various registries. Wow. V6 is getting out there, look at all those allocations. This is what the BGP announcements are. Notice that this ends really last year and Geoff' data shows some uptake in the last ten months to a year. So that is cheering, that is not shown here, sorry these are old data. But this is not a pretty picture. OK. But you have already been hearing this story today.

So, last year is a little better. We need the run out to be somewhat optimal and somewhat fair. We are going to have a train wreck but maybe we should put on some seatbelt, pass out a little food. My wife used to say think of all the people on the Titanic who didn't eat dessert so. Are the current administrative mechanisms fair? What is fair? OK. Is this fair? This is ARIN, the last couple years, all the addresses, 83 percent went to 24 organisations. Is that fair? Don't know. Other regions' distributions are somewhat different because of different policies, etc., and there is a proposal well, we won't go there. It markets the model concentration in north America but meanwhile a newcomer at the low end may not be able to justify getting a 20 to a 24. There is a barrier to entry at the low end. OK. Replace that barrier there very intentionally. I am not sure it's fair. Is this how we think the last few /8s should be distributed? When we get down to the end we are going to give them all to five big telcos and the hundreds of small people trying to get into the market are blocked? OK. Why we do this? Incumbents, people who were in the market were saving their routers, they are our routers, right? We say we are saving our capital costs at the expense of a barrier to entry. Should we be doing this at the end? Instead, giving me some tools to deal with the idiots who are deaggregating, right. And you will extend my routing life well past the end of v4 runout. OK.

What shouldn't we do? Or what should we do. I am not an expert, I admit it at least. So it's a differentiator. Even distribution of RIRs of the last /8s, the proposals before us, I think it's a very good idea. The reason is very simple: It's not to be fair on allocation etc., etc., it's to give each region the ability to plan the last ten meters. Just it's a planning tool. Within the RIRs damp big requests or requesters. Don't give the last one or two /8s to three companies. Enable small requests. Maybe even notice this Word is "might," just some ideas. Enable saving of the last 16 or so or two 16s or whatever, in reach region for things we don't know about. I am a little old and the list of things I don't know about gross daily, OK? And for the remaining (grows) to get better use out of IPv4 and open market with transparency. There is space to be gotten out of v4. This is in the ARIN region. These are what we call legacy and historic prefixes. What I have measured here is I took the prefixes that were allocated. A graduate student called Ying Wing Zang is responsible for this and deserves the credit. And she is not here to answer questions. So, the of the v6 legacy space, 50odd percent has no match at all. In other words, no space in that prefix is announced on the public Internet. Now, the Department of Defence of the United States government has a large bit of it, but don't ask because they will bomb you, too. In the pail blue percentage of the cases, what is announced (pale) is part of the prefix, not all of it. Maybe it's fragmented into two or three. The yellow is a sub net, so the what was allocated what is being announced is actually an aggregation of what was allocated maybe two chunks, etc.. and the blue is a perfect match. But this is the number of prefixes, this doesn't give this is the number of customers, this isn't how much they are spending. We have some idea that have when we turn it into /24 equivalents. And this shows you how those /24 equivalents are spread over the v6 the prefix space. So, even though there are some /8s, the amount to be recovered from there in legacy is actually not large. The weight is out here. That is legacy space.

There is also a lot of underutilised space that is since the registry started allocating, post-legacy, that they have given out space, rightly under policy, etc., etc., and the recipient hasn't used it as well hasn't utilised it fully yet. How do we put that v4 excess space to the best use? I choose this funny little term on purpose, because the lately in the last year I have been hanging around with strange people like economists who decided what we are doing is interesting to them, and so they have come to listen to us and they have actually come to talk and I have learned a bit, not much, but I keep trying, and they tell us the best use is supposedly what markets do, clearly they are capitalists. But essentially that if we have a market in it that the best use is found by people willing to pay for it.

AUDIENCE: Show me the money.

RANDY BUSH: OK. There already is a black market in v4 space, whether we like it or not. Would you rather have a black market or an open market? That is the choice we face today. I personally prefer a flawed open market to amateur overregulators. Does anybody here read the ARIN PPL list? You will know what I mean. It is just insane what people are doing. And we have a wonderful proposal here to give away v6 space with every cracker jack box, right. And (cracker) and we will look at that ten years from now like we look at the /8s in v4 space, right? Who was crazy enough to give that away. So how do we make it transparent and safe? The first problem is the buyer needs assurance that the seller can actually give it the title. Do they have the right to sell it, do they actually possess it? So the who is date day no formal means of verifying if a customer holds IP space, no formal verification of routing announcements. So we want to be able to verify the assertion of the rights in IP space, we want to verify rights of ASNs to originate prefixes back to the routing problem. We want the correct routing announcements and know that they can transfer it to us. OK. The answer, as you heard Trudy and others present yesterday, is a formal informal in the cryptographic sense, verifiable public PKI. OK. The goal is to handle both resource ownership of ASNs and IP space and verifiable transactions with others of allocation, sub delegation, transfer, sale, lease, etc., etc. the one that is being built is distributed and when it comes time for routing you will be able to put replicas down by your routers. PKI architecture, allows you are going to be compatible across RIRs but each RIR can have their own business processes, etc., and like the IRR, if you wish 30 or so or 50 ISPs will probably run their own processes. It kind of looks like this, this is the one ARIN is building, which is this is the current back end, the back end talks to the engine this way so this is a private protocol but our engine talks to your engine and talks to IANA and talks to its customers this way and there is a publication protocol, so this we refer to as you the up down protocol. This is the publication protocol and these are the things that are going to be public and are being documented. OK.

APNIC did a simple prototype. ARIN pulled it in open SSL so ARIN and APNIC are kind of driving the protocols and design models. The RIRs are all working on it and the results will be open source.

So, we can't make out more IPv4 space and we are not going to fix the speed of light, either. What we can do is allow markets and tradings to give us the best use of the remaining IPv4 space and we can see that the customers are safe in doing so. OK. You don't want the lawsuit that says "he sold that to me for ten million dollars and he didn't have this and your whois data said he did." So these are the nice people who have supported all sorts of work and if you have any questions, Mark Dranse will answer them.

(Applause)

CHAIR: Thank you, Randy. Any questions for Randy?

TOM VEST: Tom Vest, amateur over regulator. Randy it's interesting; you suggest that the distribution now is unfair. In point of fact every industry on earth is characterised by that kind of distribution, so it's if we are unfair we are unfair in exactly the same way that every other industry on earth in human history is unfair, but I am curious about your assumptions.

RANDY BUSH: Tell me the other industries that have is a barrier to entry at the end that is administratively placed there by the current incumbents and have gotten away with it for very long before the lawyers came in?

AUDIENCE: It's probably every other industry also. But actually that is my question, since that was just a statement; my question is about, again, you believe that the system has been unfair today. However, your market model you are suggesting in that model, the people that you think are unfair are going to be the sellers of the space. Now, how many people here in the room think that v4 address space should be sold at a price that you know whatever the market will bear? Either a showing of hands, the price should be whatever the market will bear. How many people think the price should be capped at some rate. How many people have no opinion at all? Again what makes you think that the market price whatever the market will bear for a nonsubstitutable unique resource of which there will never be any more of, is going to be is going to yield a distribution which is more fair than the one we have now? I mean, I think you can't actually advocate markets and assume that is going to happen. I mean you can't believe it's going to happen, because markets

RANDY BUSH: I will work that out with my shrink.

AUDIENCE: Consistency I guess

AUDIENCE: That is difficult to do. So Randy, would you actually then be in favour of a policy proposal that dropped the minimum allocation size to 32? So that small bits and pieces

RANDY BUSH: First of all you are talking to the only person in v4 space who ever received a 33. Secondly you are talking to the person that proposed radically dropping it in what was it eight years ago at the Denver ARIN meeting and got the hell beat out of him by the amateur regulators.

AUDIENCE: Right. So you are still in favour of something like that?

RANDY BUSH: I believe that we are going to see allocation I don't think you are going to get policy change before v4 run out that slices things smaller than a 24. OK. I don't think that is going to happen knowing the momentum of our industry. But I do believe that I think letting it go down to 24 sincerely not unreasonable.

AUDIENCE: To what extent would you count the professional regulators into your forecast in terms of there will be a market for v4 addresses and so forth?

RANDY BUSH: Could I count I didn't see any forecasts, but I believe they are professional regulators trying to understand and study this now and I am trying to learn from them. The concerns I am hearing are not the ones of the left end of the curve that I showed, you know the big people getting the big chunks but there are concerns about that we have, through our regulation, play the incumbents have placed intention barrier of entry to the small folk.

AUDIENCE: So would you say at the end of the day, the current distribution model for IP addresses might not work as we know it?

RANDY BUSH: It certainly works as we know it. Whether it works the way that the civil society thinks it should work. That is less clear. And those folk from civil society are wondering over and trying to understand it and and I learned from talking to them and come join the conversation.

AUDIENCE: We have had what is basically called a cat goerical distribution thing so needs based allocation model to date. If you have certain assets that you can credibly claim to have control of, you can get allocations of address space. Now, that is some kind of mechanism, that is some kind of definition of fair that you actually have some means to put address space to good productive use. Now, what a market model would do is substitute that categorical eligibility for the highest bidder will win.

RANDY BUSH: Actually if you look at the proposals in the regions, you will see that the current proposals say that the recipient of a transfer has to qualify just as they do today. I am not saying that is good or bad, but I think you are picking an extreme along a continuum

AUDIENCE: I think you are assuming that the if we go to a bilateral decentralised trading model, that people will still want to have some interaction with an institution that sets rules that you think are unfair, and I don't see why that how that would be consistent, either.

RANDY BUSH: As I said, I think you are picking extremes on a continuum.

CHAIR: Sorry to have to interrupt. I think we are going to run out of the coffee break time.

AUDIENCE: I would like to know what you are thinking is on what impact, if any, this has for IPv6 allocation policies.

RANDY BUSH: OK. I mentioned I don't think v4 trading has a serious impact on IPv6 allocation policy. I think I think the IPv6 allocation policies that we are doing today and that are being proposed, are unwise a long a number of dimensions; one dimension is when the registries don't have any v4 space left to give out that they got for free, and they are trying to figure out how to get their next meal to survive in their enormous bureaucracies, the fact they are trying to give it away with every cracker jack box does not bode well for their income statements. I think that in the long run we will view some of the policies in giving away big space that we are doing in v6 to market it as reproducing the swamp and the big legacy space we have in v4. Those of us who are unfortunately a little old, we have been to this movie. It was bad, and the popcorn was soggy

AUDIENCE: I guess part of my point is you know, what can we collectively do to impart stay focus on making sure v6 works better. What do we do with the end of v4

RANDY BUSH: I think that is a really good point, is we are arguing about a decline in quality, well it's the actual number of we are arguing about the end of that. We are in fact where in fact we should be looking and focusing on how we can make v6 rational early in the game. Point well taken.

CHAIR: Thank you. Just before the coffee break.

(Applause)

JAMES ALDRIDGE: Going back to the graphs I showed at the end of my talk over the 40 minutes or so, we have lost almost all the OK. We actually some of the access points reloaded themselves as we were reconfiguring or just after, looks like they can't cope with the sudden change of settings. But we have got just around 110, 120 users on v6 only networks at the moments traffic is around 1.3 megabits per second. And the CPU load is around 8 percent, and I will include more graphs for the full v6 hour in my talk at the v6 working group.

CHAIR: Excellent. Thank you very much. This is the last of this session. Thank you very much to all the speakers, to the RIPE NCC staff who made this whole thing work and the other volunteers who helped. We are now into we are into the coffee break. I haven't talked to the working group chair of the next meeting I suggest we suggest are back at five past 11. So we have sometime out there. Thank you.

(Coffee break)