These are unedited transcripts and may contain errors.

RIPE 67, Athens.
Plenary Session 2 p.m.
14 October 2013.

ROB BLOKZIJL: Good afternoon. It's two o'clock and we will start the 67th RIPE Meeting. If people can find their seats. Welcome to Athens, my name is Rob Blokzijl, I am the Chairman of RIPE. Before we start our Plenary programme, I want to say a few words. In the first place, welcome to this RIPE meeting in Athens. It's the first time we have a RIPE meeting in Athens. It's the second time we have a RIPE meeting in Greece.

I have a few technical announcements and then our host of this meeting, I will introduce you to the host and the host will say a few words as well.

In the first place, my friends and colleagues from the RIPE NCC told me that our local host, the GRNET, have been already doing an extremely good job in the preparation of this meeting. But they are not the only ones involved; we have more sponsors later in the week: NetNod, CCC?IX, AMS?IX, Compass?EOS and Fortinet, all have contributed to the various aspects of the meeting this week, so we thank them all.

Secondly, you may have noticed there is an experimental IPv6 only wireless network in the air. This is a pilot project of the IPv6 Working Group. So it had nothing to do with the RIPE NCC provision of facilities. You are ininviteed to play with it, but don't complain to the NCC, complain to the RIPE IPv6 Working Group. You can, for instance, try Jan, or if things don't work and you think it smells of IPv6, complaints to Jan.

Tonight, I want to remind you there are two social events taking place in the same room. One after the other. At half past six, drinks with the RIPE NCC Executive Board, that sounds very formal, but it's an opportunity to find them in one place so you don't have to hunt them down if you want to talk to them.

Followed in half an hour later by a welcome drink for this whole meeting.
Last but not least, on Thursday, we have the RIPE dinner, there are still some tickets available, €50 each, and if you are interested and you don't have your ticket yet, the next coffee break, go to the registration desk.

Now, it's migrate pleasure to introduce to you Yannis Mitsos, head of operations at GRNET who is our local host. Yannis.

YANNIS MITSOS: Good afternoon everybody and welcome to Athens and RIPE 67. I work for GRNET, I am head of network operation centre.

GRNET stands for Greek Research and Technology Network. It's a not?for?profit public organisation that serves the Greek research and educational community in Greece, offering mainly networking services and now ate days we are also started slowly but aggressively to offer cloud service to say our community.

A few more words about GRNET. It works under the auspices of the Greek Ministry of Education. It's a company with 60 or 70 employees. And you may find similar organisations all around Europe and around the globe. Those organisations discuss RENs national research and educational networks who more or less serve the name objectives and goal, namely they offer advance networking services for their communities. Having said that, there is also the strong support of the European Commission which has ?? which subsidises actually a big network, it's like a parallel universe. You see the network here in the picture. It's a huge network which is interconnected as well as with the commercial world, as well as with other continents like the United States, in Asia we have big links trans?connecting our stations with them.

Area those NRNs important. They are important because they pave the way for the innovation. To give you some examples, I have prepared a few slides starting from the DSDN examples. You may see here a case of a test net as a service that is being deployed in the previous GR network and this has a case of facilitating the exchange of huge amount of data that are produced. So, they will offer the service in the next year to the researchers in Europe, distributing the data created in these big experiments.

If we come a little bit to our micro world for GRNET. GRNET is also very keen in putting in the production of innovative services. Actually a year ago we have rolled out a service that is used firewall on demand. Why is this important? This is important because it utilises BGP employee spec. It's a protocol that has been introduced back in 2009. Cannot find ?? cannot meet many production grade installments around Europe. However, this has been part of our portfolio, the last year, which also gives us some preference compared to other solutions.

With regard to the DNSSEC deployment. GRNET has a service that provides cheap connectivity to the students in Greece. Since two years ago, if I recall correctly, we have enabled DNSSEC in our name servers and you see some results that have been published recently, last July if I recall correctly, stating that the Greek student service ranks third in the world for DNSSEC deployment. Also, 98 percent of the user base are always utilising DNSSEC. This is a great achievement and now we are trying to put it into production for all of our network as well.

Last but not least as I mentioned before, NRNNs used to look after only network services in the past. Nowadays, it's the trend, we are obliged, let's say, to move on to the cloud services; therefore, we have developed from scratch a cloud service infrastructure service that is offered to the Greek research educational community. It is being now ?? we are starting now rolling out better version. Of course, it is full IPv6?enabled. And more the day after tomorrow you will be able, for those who are interested in this service to have a small presentation in the room which is right after the registration desk, so feel free to visit my colleagues there, there is going to be a small presentation by the capabilities and the features of this cloud service.

And last but not least, just I put this slide emphasising that the need for federated activities all around Europe. We are progressing into different ways here. One way is the ?? we have identified actually the research community in Europe has identified the need for mobility and, therefore, the service, it's called the drum has been introduced many years ago, so students or researchers from one country may travel to other country where it's enabled and connect the network without any nasty things, and there is also in parallel of this, a single sign on mechanism in place, our cloud service, all of our new services rely on this single sign on mechanism. We don't have to maintain now local password user names, things like that. Everything is delegated to present the providers of our network.

So that's a short version of our presentation I was asked to put very quickly for ten minutes. If there are any questions...

Thank you very much.


ROB BLOKZIJL: This concludes my short opening session and now I hand over the hammer to Filiz, who will take the rest of this session.

FILIZ YILMAZ: Good afternoon everyone, my name is Filiz Yilmaz and we will start the Plenary Sessions from now on together with my colleagues on the Programme Committee. But, the very first start of this will be an update from us as a Programme Committee.

This is what we do. Just a short introduction, not even an introduction; just a refreshment of what we actually do as a Programme Committee. We are responsible for the content for the Plenary Sessions which are also including now the tutorials, workshops and BoFs. So these are laid out on Monday, Tuesday, entire Tuesday and we use half of the Friday too for that programme. As you can see, these are called Plenaries on the programme meeting plan or they are marked as tutorials, workshops and BoFs. So those you see on the convenient are the content that we are responsible for. So if you don't like what you see on those days, blame us, come and talk to us. The days, Wednesdays and Thursdays, are mainly for the Working Group sessions and those are ?? the content of those are determined by the Working Group Chairs, okay. This is mainly ?? sometimes we receive some confusion in regards to, you know, who decides on which kind of content for which part of the meeting. I hope this clarifies some of those concerns.

So, when we are planning the content of the Plenary, there are certain things that we keep an eye on. This is mainly, we want to make sure that the content of the Plenary Sessions are appealing to the general audience of the RIPE meetings, which is mainly, in very broad terms it's network operators and operational community and while doing that, we obviously need to keep an eye on the little balancing act, which is basically finding the fine tune between engineering content or technical content and the policy content. Nowadays, as we see, there is a lot of collaboration work coming in also relating to engineering work, and the border lines or the borders are not that clear any more. We do our best. Again, if you don't like it, come and talk to us. This is the best way of getting feedback from you, although we are having some other mechanisms, too, which I'm going to show you in a minute.

The contents of these sessions are to be complementing to the RIPE Working Group agendas, obviously. As you can see, you know, almost for every subject, operational subject, DNS, routing or address management, there is a Working Group as well, corresponding Working Group. The main thing is they really go in deep and produce work around these subjects. While with the Plenary content, as we said, it needs to be appealing to a wider audience, so, they are shorter presentations as well, that's the other difference. You get 30 minutes here and we recommend obviously leaving some discussion time.

Why do we do this? Well, obviously there is that motivation part of the operational side of the RIPE meetings. And we want to encourage, hopefully, by bringing good content to the meetings, motivate attendance and it will be even better and better. Last week's meeting was very good in that sense. Let's see how it goes. I don't have the numbers yet. I guess we will see them at the end. I hope the interest continues through good content.

Finally, obviously there is education part, and the entertaining part. We don't want to bore you with bad content, so shows are the other things we considered.

Now, these are the people that are currently on the Programme Committee. We have a full house, all 12 seats are occupied at the moment. However, one is coming up free, and there will be elections on Friday, so if you are interested to join this group, please send a mail to pc [at] ripe [dot] net and a bio and statement of interest, and a picture please, so that people can recognise you. Two people joined this group at the last RIPE Meeting and those were ?? one of them is Mike Hughes, who has been contributing since RIPE 66, and the other one is our local host, Andreas Polyrakis ? he has contributed to the programme for the local host site. And for the elections, to sum it up: if you are a candidate, if you are interested, send your information to us so we consider you as a candidate, and for those who will need to vote during this process, you will need to have logged in with your RIPE NCC account. If you don't have one, please create one already.

Now, before you jump on the board, also we want to make sure that you take a moment to reflect what you are getting yourself into. There are benefits and expectations as well. The expectations are basically ?? this is about good talks, good speakers, we bring content in so part it have is goes out and chasing people and that takes a lot of time, believe me. And once we receive proposals, your task will be to read through this content and rate them. On average, you know, the RIPE meetings differ from summer to autumn meetings, the proposals, number of proposals we get by as between 40 to 70 submissions you will have to look and rate and we expect you to attend three or four conference calls during the period of preparation for a RIPE Meeting, which means like, in three or four months, you will have to have three or four conference calls.

And you need to be constantly in communication with the committee, with the presenters, with the committee, with the RIPE NCC staff, so this is what the expectations are. But when it comes to benefits, it's satisfaction; basically, you get satisfaction serving the community. You bring something, your expertise and contacts. Through all that, you bring value to the RIPE Meeting through the content of the Plenary Sessions and you get a bit of industry and peer visibility, obviously.

But, that's pretty much it. There is no other benefits if ?? you know, let's set the expectations straight here.

What we have been busy is that 12 people recently since the RIPE Meeting, last RIPE Meeting is, we worked on a bit more together with the RIPE NCC on the submission system. As you may remember, there was a change. So if you are still experiencing some ?? or if you have feedback about the submission system, again please come and talk to us.

We introduced a new category, which is called workshops. This normally ?? before that we had the tutorials and the BoFs, as those sessions where people kind of bring in demos to the table. But they didn't really work out well in the sense that they weren't really hands?on workshop settings, so we thought we'll throw this at the table, and we have one already with the RIPE Labs, I believe, it is about, and ?? so, go and see how it goes this time and let us know if that works as well.

Through that, we updated our call for presentations. This was also a demand for information. It turns out that what the Plenary talk versus tutorial versus lightning talk wasn't very clear, and to we put more effort in explaining that. I hope they get noticed. These are the links. Please go and have a look on our categories and let us know again if you have further feedback.

Some stats: We received for this one, for this meeting, 34 Plenary proposals, together with 5 BoF, 3 tutorial and 2 workshops. Lightning talks were popular this time. When I prepared these slides there was 10, now there is 13 or 14 of them, something like like that. And on the 23rd August, we published the draft agenda. As you may remember from the previous updates, we really tried to publish the draft agenda two months before the RIPE Meeting start date. We do that because we received lots of feedback in the previous meetings that if the agenda is not out there, people are not well?informed if they want to attend or not; also, what is going to be discussed, etc. Bear with us. Some of the fall meetings, which is after the summer school holidays on the northern hemisphere area, say, is a challenge in getting content, so this time we were a bit delayed, it went down to you know, one?and?a?half months, not two months on the mark, and then on 17th September we came up with the final agenda.

So if you still have concerns about this timing, let us know again. We also hear interestingly enough it's too early from some people. We like, to you know, have a shorter period. So it's a balancing act because people, attendees would like to see what's on the agenda, while proposers want to bring the latest content so they want to submit as close to the meeting. So... yeah... it's an ongoing battle.

If you still want some mike time, you can still get a lightning talk time. The popular ones, as I mentioned before, it's the ten minutes talk with five minutes you talking and comment five minutes before discussion time for Q&A and you can still submit them. We will decide on the slots for tomorrow before half past three, so you may, if you submit something now, you may be running up for those, for that election process. If not, it will be for the Friday slots, which is one or two.

Now, how can you help a little to your PC members? You can rate the talks. On the Plenary agenda, you will see every presentation will have a link to be watered and you can water quality of the talk in terms of its content and if you liked it or not, how you found it relevant, and if you do that, RIPE NCC folks are kindly offering you a prize after, you know, a draw. It's an Amazon cheque for €100 ?? so go for it. And if you also ?? if you have more specific comments, we are all out here, come talk to one of us.

So, this is all I would like to say, and have a good meeting, and I will leave it now to my colleague Jan to continue with the rest of the session.

JAN ZORZ: Hello, my name is Jan and I will take care of the housekeeping for the rest of this opening Plenary Session.

Welcome to Athens. Good to see you all here back at the RIPE Meeting. And our next presenter probably needs no special introduction. When I thought that it's really cool to have three gigs of connectivity to your home, then I learned from Peter that his mum is somewhere in the north of Sweden, has 100?gig link to his house. Wow, that is cool. And we are really, really happy to have Peter Lothberg here. He will talk about the TeraStream network and IPv6?only transport and how they did it. This is ?? we, as a Programme Committee, thought this this is such a good solid content that we gave Peter 45 minutes, so this will be a bit different from the rest. Peter...

PETER LOTHBERG: Thank you. I am very happy to be back at RIPE. It's growing still.

I'm going to talk about how you can build a network, and basically, I have been building networks for half of my life, and before that I was still trying to build networks. I started off talking at RIPE meetings, and at that time we were making presentations on stencils and fold papers and pens and God knows what, a long time ago. And now, we are moving forward and we have all these fantastic tools and everything, and I said, okay, let's see what we can do with this.

There is a word in my presentation that doesn't really exist. It's called key IPv6 tunnel. Basically, we came up ?? we took a technology that we had with IPv4 and put it in IPv6 so you can carry an ethernet frame. We will all ?? it P.V.?free, but we decided to pick out a name. So if you are looking after the draft document which describes what we are doing it's called the keyed IPv6 tunnel.

So, a couple of years ago, I was asked could you come up with a proposal on how to build a network for the future? And a network where basically we can deliver everything at a low cost at very high performance and at automated as possible. And I sat down and scratched my head. My first answer is, you are a phone company, no way. Then I go, maybe there is a chance we can actually make something better and I sat down and I thought, okay, if I can make a proposal and start from scratch and propose whatever I want, I'll try to do this and see what happens.

So I sat down and scratched my head and said, how does a network look? What do we actually need? And the first thing was, you look at an existing network. There are boxes. You may look at an existing network and ask that organisation for their org chart. They usually look the same. So, try to put a few boxes as possible in a network. That's kind of hard, but you have to put your finger on something and say is this doing anything the customer really cares about or does it even know that it's there? If the answer is it doesn't do anything or the customer doesn't know anything, take it out.

We want to have as few interfaces as possible. Basically, a lot of people are building a network with routers back?to?back or boxes back?to?back, and I go, look, there are no paying customer, let's get rid of them. Let's try to have as few interfaces as possible where they don't face something that does something, okay.

Well, another thing we tend to do was we put all sorts of features in the routers, you have an edge router, I think it's called a P this day and you put VPN, all sorts of things in there. I go, let's try to not do that. A ping master 2000 was the old phrase. Don't put any services in the network. The service of the network is to deliver packets. And the service is deliver a packet at the given quality, at the given time and nothing else. Like, everybody goes, but what's your service? Well, look, you have power, you have electricity in your house, nobody ?? everybody wants the electricity feed to be reliable, they want it to be safe, they want it it work all the time. Nobody goes to you and say, you need that sexy TV we saw yesterday; it's just a utility. So it's very hard sometimes to tell people the plumbing and the backbone that actually holds this together is actually very important, and we are going to see at the end of my talk that what happens in the home network maybe actually where we have the problems today.

You build a network with IPv6 only. Now, I don't like IPv6. That's probably most of you know it has a lot of broken things in it and I think that if I were to propose something new it would never happen so. Okay, let's take and fix what's not really working the best and see what we can do.

So, I said okay let's build a network that only speaks IPv6. Use IPv6 for everything and don't do any layer violation. And use layer?free tools. Ignore what the transport is because it will change over time. And in order to be in the market, we said okay if it's not an IPv6 packet, what is it? Well, it's an ethernet frame and what's in that frame we don't care about. So basically, this network has one service. It can carry ethernet frames in IPv6 packets, use of the key tunnel we talked about earlier.

So, IPv4 is a service in this network. It's done by the home gateway or software in your end terminal. Layer?free VPN is a service, it just doesn't do anything.

Then you always run into, oops, this doesn't exist, Peter, you can't do that or you can't ?? it's impossible. So we put that on the list of things impossible. Let's go fix it. There is a bunch of things you have to have to build this, and we said, okay, let's go make them happen. There is a lot of parallel activities of doing things that are missing.

Now, if you want to deliver packets cheap, you want it to be fully automated. How do you build a network that runs by itself? How do you build a network that you don't need to configure by hand? How do you operate and manage that? It turns out that the other thing you want to do is talk to other operators. You are not alone in the world. Your network needs to talk to another network on administrative level. It's called network to network interfaces. Provide a way for the network management to talk to somebody he will else's network management so they can use the resources out of your network that they are using or not that they have full visibility and control over that they are assigned to use.

We need to put policy past the network into the home network because the home network may be more complicated in the future. And I was struggling trying to find the technology and I think the easiest way to do this is, we need to solve the multi?homing problem of home, the handset you walk around with is multihome. It speaks 3G, 4G and Wi?Fi. So let's solve the multi?homing policy problem. It's actually more or less trying to soft multi?homing policy problem. Let's try to work on that and use that as a tool, so if my network has multiple services inside. I am three providers or two, whatever, that's a generic problem we need to solve anyway. It's not specific to what I'm doing and I think it makes the Internet a better place.

At last we are trying to give the customer the possibility of using the network. Okay, here is a /56. I think that is still your policy, I hope. Here is 8 bits that you can use inside your network so you can build something. That gives the customer 2 to the 72 end nodes. This doesn't fit in anybody's house. It doesn't even fit on the planet unless you want to make it bigger.

So, let's try to put this all together.

Cost is going up. Revenue is going down. They have all sorts of complicated things and they think the more boxes they have the better network they have.

So I said okay, what do we actually need? Remember, this is 20 years ago. What do we actually need? We probably need this. I couldn't live without that and I think there is another bubble here which comes out of how you configure things because I think this is kind of not the way you configure the network in the home in the end because it's a network as complicated as the backbone. But we ended up saying IPv6, we have tunnels, which speaks big bit ethernet, we have what is called and OSS gateway. In the old days you had network and OSS and a business system and all of those things. We said, look, put a slim layer on top of the network and make nobody talk to the router. Talk to the gateway that talks to the routers and make it the data model driven thing so you tell the network what you want it to do, not how to do it.

Now, after I did this, the world became buzz word bingo. So it's called SDN and all sorts of things. Now, I'm not trying to control anything in the router. The routing system of this network is 1995 standard IGP, IBGP. It doesn't do anything sexy. The only thing I'm trying to control is the services, not the network. Restoring traffic. Putting the right packets on the right path, is still done by an old traditional routing system. It's not a service ?? it's not an SDN, it's a service defined ?? it's a service defined service. Right.

If you make a drawing ?? I stole those papers from the printer ?? you end up saying okay I have an edge device. The edge device does the service. There is something between the edge device and the customer access maybe. It turns out if you remove. There is all sorts of bus words here. It turns out the smartest thing to do is probably put GigE on the [coarx] cable, but this is the function ?? this is the box that I end up calling R 1 because that was first one. Here is where you do all the services to the customer. The rest here is just transport and make sure you don't over subscribe it. I think you can push this down all the way to suite cabinet and that way you do 100 gig down to the street cabinet. Then do you IPv6 from there. You need to build a network so I put these boxes here called R 2. This is the only other device I have. If you only have edge devices. This one and this one, you don't need the labels which something here because then this becomes more complicated. So all the devices are edged. Half of the traffic is going to the outside world so I put peering everywhere. 40% is something that you can put in a data centre. Let's put the data next to this box. Put connected servers directly to this box. We are going to use the server as much as we can to feed stuff to the customer. 10% stays in nigh network. I think that's a high number.

I want the network to be fully redundant so basically all of routers are multi?homed to two of the back?boned boxes and using ISIS it's not over?subscribed. I have 1 plus 1. I have full amount capacity as on the right side as I have on the left side.

It looks like this. We have service, we have the customer and we are moving packets. There are data centres and we have the rest of the word. This picture wasn't made for you guys. It was made for the marketing people.

How do we do this? Well, so I said, take my inner packet thing and turn it into principles. We'll use a number of technology. No nothing, see you. Use IPv6 for doing everything. So I'm using IPv6 for everything. Not R2s in the data centre, still needs to talk to the old Internet, because that's where most things are but hopefully that will go away and by using general purpose hardware for function in the network I'm hoping that the stuff I need today to get boot?strapped transferring IPv4 into the IPv6 packets encapsulating could be used for something later on when traffic shifts to be a native IPv6.

I do the IPv6?based ethernet service and it's even worse, I did ?? I went and read the met owe ethernet forum which said all the open nation maintenance stuff, all this thing they have to do look backs and pings whatever we actually support, but it goes in an IP packet, we don't really tell them.

No interfaces back?to?back. Better face either goes to a server or it goes to a customer, it goes to appear.

There is no need for a transport network. If people get off their, I want to do things differently and standardised, there is a market for making optics and silicone. It turns out that you can probably make 100 gig transceiver for eight hundred kilometres for a thousand bucks. So why not just assume that that's going to happen and put all the optical stuff, all the transport thing on the router itself. The router you need because it actually does the service the customer is paying for.

Converts the whole thing, use packets. Here comes I didn't wanted to have this talk. I amen coding the service and the ?? I'm telling what the traffic is in the IP address. I am encoding what this packet is doing, where it's going to, where it's coming from in the IPv6 address so. I'm not touching anything in the packet. If the customer has [TOS] bits, ICM bits, whatever they have. I'm not touching them. Actually at the end of the network I am honouring them, so if they can get the sender to send the right bit combinations and they are using more bandwidth than they are buying from the service provider, they can customise packets the way they want. I'm going to go back and talk a little bit about what we did and how. Now you understand probably this is a way, I think this is the right thing to do because it makes it possible to build a better and bigger Internet.

So, summarising the design of this thing. Start over here. Here is some kind of access, forget all this fancy things. This is an ethernet switch to the customer. Logical interface here that one per customer, here is where we configure all the servers. Everything else is just moving packets. The data sent certificate sitting on each side of this, and the data centre contains the content, the DNS server, the configuration server, whatever, and the net boxing, whatever we need. The RT basically used packets. We plug wholesale if you do that, if you plug the rest of the Internet and hopefully, you know, 100 gig is not very much bandwidth, so hopefully those interfaces, they are coloured, so we run one fibre to our peer and put 60 or 80 wavelengths or something on that. This is the entire network. It's IPv6 here. It's IPv4 over Softwire, there are a gazillion ways of doing Softwire and basically it's a fact between the deal and the end point and something that runs in the PC makes it possible to play around with multiple versions of it. If it's not IP I take the ethernet frame and I put it in an IPv6 packet and I can send it somewhere. I can send it to November negligent, decapsulate it send it out. I can send it to a server and do something with it; I can send it to the other end of the world but I can provide any kind of service that is not IPv6, so if you want to do VPNs or something, we will use that.

So if you don't want to put labels and you don't have anything to put your packet inside. You need the packet by itself has to tell what it's doing, and what I ended up doing was, I put information in the address here that says here is what the packet is doing. So, basically, the service type ends up being 3 bits. The bit is a that I say this is internal to my network, the thing that states it's infrastructure versus customer, the bit here is something which I want to hit people like Nokia and Ericsson a bit later. Then all of this basically says what port and router it is. And then this is the slash, this is the 8 bits we are giving to the customer and the lower 60 bits all go into the customer land. But by looking at the packet, you can say, this is from the customer talking to the Internet, this is from the customer talking to the voice network. So I actually hand the customer free blocks of /56s, one for their Internet service, one for TV service and one for phone service. I am basically 3 ISPs inside the network, that's how I do the policy and take care of the net neutrality and all of that stuff, because which one do they apply to and then we follow the rules for whatever that service has to have.

So, how do you build this in practice? Well, you take a router. And you take a piece of fibre, and you run the fibre from here. This is a pair of fibres, we ended up calling this a horseshoe. It's a half rim. Then say we need two routers here. Every router here is good. For the way we did is every router is good for 5,000 customers. For supporting ?? when I was counting, I did this. I took 40 megabits by directional from the customer. So the customer is 40 megabits 24/7. If you have 200 gigabits, as you have here, you divide it by 40, you get 5,000. That was a defining criteria. Most people, when they build DSL networks or things, they use 40 kilobits. They haven't seen the 8k or 4k TV. So the router here has all the optics on it. This is, for example, the working path. This is the backup path. The backup data centre for here is actually not here, but this takes half the led and you will see a dirty trick we use to do backup once one of them fails and this fibre here basically carries the traffic. So we are building point?to?point links. So try to make a drawing. This is one side of the horseshoe. Here is all the boxes that sits on the horseshoe. This is the 10% circuit and 10% out of 40 megabits times 24 million users becomes a lot of capacity. This is basically 10 parallel 100 gigs, from this box to every other box. So if you look at the graph of the network, you basically can go ?? you have two ways of getting from this box to this box. Either you go from here on the direct next hop on the backbone. If the backbone is broken, you have all those boxes in parallel so you basically see 60 or 70 parallel paths from this guy to this guy so you can do transit. And remember, I'm only using half the capacity on the fibre, on the horseshoe, because I am counting on it will break. So I have enough capacity here to carry everything so this side may be actually idling, or it turns out we have basically 50:50 load.

So I got a pretty redundant network. If you try to plot it out, and all the backbone, R2 to R2 boxes aren't there because the picture becomes unreadable. Here sits a router that tries to talk to another router, so customer sits here. There is an R1 sitting on this horseshoe. It may want to talk to a guy sitting on this horseshoe. Well, he can go this way on the backbone to this side, come down, or he can go that side or he can go any combination of that. Our favourite software to define what happens in failure scenarios crashed. When you try to figure out what happens with the second order failures or third order failure, they became, it said poof, I need to buy a bigger server.

Now, if this data centre here fails, I only have half the capacity to support my customers. So I have ?? if this data centre fails, I fail this data centre too. Because I have 18 more data centres where I can use a little bit of them with load sharing, so when I lose one data centre I take this out and I use the rest of them as backup because I'm assuming the network will still be there. And, of course, if you end up in a scenario that was second or that was second order problems, I'll use whatever I have.

I invented a new thing: I called it service guarantee architect. If you, today, are connected to a service, whatever you have, it has certain characteristics, they probably built that service by over?subscribing everything from the backbone everywhere so the service quality is whatever you get. When the neighbour is using your network it runs slower. They are using a worldwide network to generate the service you are getting and this doesn't work because you have no way of figuring out what's actually going to happen if you really need to. People talk about QS, blah, blah, blah, blah, that is only who are you going to knock off. So it doesn't help. What we ended up looking at is to say okay, we have four Qs in. I was the Internet, TV and telephone service provider and three is not dividable by two, so I have got the 4 Q. There are 4 Qs out and 4 Qs in and I implemented a bunch of functions on those Qs. Use the RIPE probe data. This is the service you are obviously delivering today. I synthesise that and configure my boxes to generate the same crap service. So I generate whatever you have today with the service guarantee architecture tools and then I let my control systems vary it over time. I completely emulate the existing service and obviously the existing service has an accepted price tag. If you want something more I have the possibility now of cranking the parameters and giving you something more if I wanted to.

We are trying to make sure that this piece of the network never loses a packet. If we are going to lose packets we are going to do it here. If you have an Internet Q, there are toss bits and ICM bits, whatever they called, in the pack packet, I'll try to honour them. So if you can say, this is more important to me than this other packet, say that you have 10 megabits coming down here and 15 megabits shows up, I'm actually trying to look at the toss bits and everything else and say this is probably what the customer wants. So if you want to patrol the Q towards you, you want to find the way to talk to the sender to mark the packets properly. And remember, I'm not using any bits in the packet, in the pay load for running the network; they are in the address.

So then the network looks like this. This is R2. R2 has peering circuits. And it has my backbone. And then I structured it so that those servers are ?? all stand at X86 servers with 10 gig ethernet. So those connections are crossover cables. There are ports on the router. There is no data centre switch, no top of the rack, no nothing. I structured it slightly this way so that external side is here, customer on the south side. They only speak IPv6. This side is facing the data centre. It only speaks IPv6. This side is facing the data centre, it speaks 4 and 6, it talks to the outside world and then I can now put my things in here. And depending on the service class and whatever, I get different bit combinations in the address so they end up having different addresses depending on what they are doing.

If you look at this carefully I laid it out so that in the future when people come to me and say we don't have routers that have 5,000 100?gig interfaces, I say let's go build up. It's a switch this way and if you make U?turns ?? what do you do if it's not an ethernet packet? One scenario is there is a build I run cables to everywhere in the building, I plug them in, they are not a customer. We need to do something to tell them go buy something. What you do is you configure the port to take everything that comes from the packet. Everything that comes from the customer, put it in IPv6 packet and send it off an an IPv6 destination address. It could be anywhere on the planet. Let's send it to a Linux box that decapsules the packet. It pretends to be IPv4, it can speak PPPUE, it can do everything. The only message is, welcome, what's your credit card number? I call you back.

So anything that is not IP, we put in an IPv6 packet and we send it somewhere. If the customer manges to enter his credit card number we will tell them that OSS, this guy needs the service and we will go configure the network.

So, everybody knows about dual stack, that's good for you. You scratch your head okay fine, we have IPv4, we turn on IPv6. Now we have IPv4 and IPv6. How do I turn off IPv4? So, you sat down and you scratched your head and say how long is this long tail? Forever. So, I think we need to ??

Dual stack is cool, how how do you get rid of IPv4? If you are turning off IPv6 why don't you take the extra step and say if you want IPv4, I speak IPv6 or IPv4 and if you are speaking IPv6 and you want IPv4 connectivity, we are going to put it over IPv6.

So, basically put something in the home that takes care of the IPv4 traffic. So, the traditional way is basically to say, I'm not saying we should use DS?Lite. It's just that's where we started because that was available standard and we could get home gateways that did it and we can get software that runs on an X86 box that did it. So that's why we did it. It goes here and it goes to the tunnel and it goes to the IPv4 world and there may or may not be a NAT box and it may do all sort of complicated things. All up to the providers to choose. If the packet shows up here as IPv4 and the guy is trying to go to Google. Why don't I make a slightly smarter home gateway they are saying you are trying to speak port 80 to Google. I can do it that way and help the guy to look like an IPv6 user because you are really want to go get rid of this IPv4 thing. And by using the standard hardware here, I am counting on what I paid initially to put NAT boxes here, I can reuse for something else. But I really need to find a way to move it over here and if some smart people shows up we can figure out a way if you want to speak IPv4 from something in your home you have a device that looks like a RIPE probe, a bump on the wire and if it tries to talk to somebody else, elsewhere on the planet. They have the correspondence box and IPv4 is point to point private thing and maybe goes away, I just don't want in my network because every time you are add ago protocol here I have operational costs and I want to get rid of pour too. This is a beautiful picture. So, my data centre stuff is all connected back?to?back with R2. I called it the cloud server data centre because one day they show up and knock on your door and say where is your firewall? Well, I don't have firewalls around the net box? Well you need firewalls. I drew this other thing and said this is the back?end data centre. That's where everything has to be the that's at least where the firewalls are.

So, here is another picture of what I just said. Home, IPv4 host, CP encapsulate encapsulate, customer?based equipment. Home gateway, plastic box, IPv4 goes in, IPv6 packet goes to the data centre, through the data centre and goes out. Piece of cake.

Now, let's build a layer?free VPN. Well, what you do is, you go and say, okay, this is the end point that needs the layer?free VPN. We configure this box to take whatever comes on that wire from the customer and put it in an IPv6 packet so now we have an ethernet frame in an IPv6 packet. So, our OSS system says we want to build a layer?free VPN service for those customers. We configure the four end points over here. And we assign them addresses of a server somewhere in the data centre, actually we don't assign them right now, we assign addresses for this end point and that server address comes out of the address mapping I showed earlier. Then we call the data centre and say, hey, data centre, data centre management, data centre orchestration, cloud orchestration, hey, guys, I want a layer?free VPN router with four ports, a console, can you bring that up? So then what they do is, they, in their open stack Linux, which brings up my little router which actually speaks all those protocols, and then they will report back their OSS system, I have started your router, here the IPv6 end point access for your four customer connections and I go back and backfill them. Where is the tunnel going? If they move something in the data centre centre, they have to tell me the end point move, but you can do this pretty quickly, and the cool thing is that you could ?? you can fire up another one of those, you can build a circuiting between them ?? logical ?? and you can start moving a tunnel one by one to the new box or you can even have load shared multiple boxing, but at the end of the day you build something that looks like a layered VPN and this uses software on this side that is all public Linux standard, it terminates the tunnels, it does everything, but it works.

So this is how you build a layer 3 VPN in this new model.

The other problem I ran into was, oops, 100 gig optics, they can't talk to each other. Router A ?? vendor A can't talk to vendor B because out of the number of ways they screw the bits on the wire, they have all done it differently. I don't know if they deliberately did that or if they did not talk to each other. We said look, this is useless. Nobody can build cheap optics because there is not enough market. Let's go make a standard. So we made a standard. Typical napkin, too much beer and we made a standard. We made a standard which says this is how do you 100 gig long haul, over 8 hundred kilometres, here is the correct code. Here is everything you need to do, go build it. And friendly people showed up and said, ah, you want a cool hard correcting code, we can give you one for free. We ended up with a standard that anybody can implement it as completely royalty?free. We go to our vendors that we have in the network that we built in Croatia and we said, okay, we want this. After the normal arm?twisting, etc., they go, okay, we'll go build it for you.

So, what we did was that we can have router A talking to router B with vendor A and B and there doesn't need to be a transport system in the middle; they can talk to each other, and hopefully our friends in the ITUT, which are meeting this week, can stop being silly and do something. I don't care what they do as long as it looks like mine.

You want to build this network. I built this network last year and not everybody has line cards with 100?gig optics on. I said, okay, it's okay if you have an external box, so we said, okay, our perfect scenario is that the optics is in the router and we talk to the router and that's fine. We use all the OSS system to do what the optical system used to do. If you don't have that on a line card, okay, go ?? do whatever you want, but when I'm talking to the router, it should look like it was that picture. So basically the people that didn't have the integrated optics, they went and found something and they made some secret talking between the router and whatever they were having, but at the end of the day it looks to me like it's on the router and that means when they finally get their stuff together I don't need to worry about, it I can just continuous it go the same way it was, because it looks like this even if it's separated. There is no way of talking to the optical piece any longer, because if you can't talk, there can't be an optical transport department.

So, the other thing I did was, okay, we are using coherent 100 gig optics, why is he not using 10 gig? That's Morse Code, 100 gig is FM radio. That doesn't require filter. If you are doing energy detection, as we are doing at 10 gig, you need the filters you only see the wavelength the transmit certificate using. If you are doing 100 gig it's like radio. You can tune the dial. If you don't need any filters why would I put filters in? So I said okay. Can we, instead of using this sexy technology, can we go buy a 25 cent optical splitter instead and just put the light together? So I did that.

So what we did was, we basically take the light that comes out of the router at the R2 site. We put all the colours together. We put amplifiers in to deal with the loss, and, when we get to one of those sites on the horseshoe, we drop off all the colours. All the colours are available everywhere but the receiver picks one. So the red goes here and the blue gets here. The draw?back of doing it like this, I have a about the DB more noise than if you have a perfect network. It's drop and waste. I'm not blocking it. But hey, if you start to do the economics. You probably have fibre, it's probably better to put more fibre in than trying to put complicated stuff here because as soon as you put complicated stuff here, somebody shows up and they want to manage it.

Now, if you have a spare transceiver here, as you are seeing all the colours, you can take your spare transceiver and you turn the light out and scan through the channels. It's called an OPM. It's a great device. If you put two channels ?? transceivers on the same wavelength, nothing works.

Doing it the way I got, I basically had service provisioning by talking to the router. I just need to make sure I don't step on myself. Power management is a piece of cake. Here is the drawback, it doesn't come for free. I am losing a little bit of capacity. I am wasting the thing, and you can actually do a DB better.

So, I was actually trying to show how it looks in a node of all the splitting but my PDF writer lost some stuff so I'm going to try and update this slide on the RIPE page later once I figure out how to show all the things in here. The light comes in, you split off some for the receive side, you add the insert that goes over here and you amplify and it goes out. Sorry about that.

It's all pretty in here.

So, the optical system for the network we built in Croatia, which is, like, 30 100?gig channels, doesn't use any optical system. This is my multi?million dollar ?? replacement for my multi?million optical system. Those are power dividers. They are the kind of ?? basically, you put the light in in this connector and you get the same light out of 42 connecters at much lower level, because you are losing energy. It's just receivers, like those.

Those are standard components that when used by the cable guys to build their optical cable network. When they are sending their RF signal down to get on your coax cable, they modulated that on the ?? they modulated the RF signal on a laser and somewhere close to the head then they basically rectify the laser light and they become an RF signal and you put in a coax cable. This is my entire 100 gig multi?channel state of the art optical system and it doesn't have any management port, as you can see.

This is six nodes that we didn't have. Croatia is a small country, and I wanted to show that this actually works with the horseshoes being at the length it needs to be. So, we basically put a bunch of simulated nodes in Iraq, so this is the six non?existing nodes. The next thing we did was, we started off building this thing. It's 100 kilometres or so. Then I learned something. It's beautiful here. Why were we building things up here? So, we got this nice city here built by the Romans and it's much prettier and they had better restaurants than over here. So we built here and here and here, so now we built this whole thing and then tire network is ?? the yellow line is about 6 hundred kilometres, carries some 16 or 1700 gig signals today and it just works, we have enough margins to basically double the distance without having any forward error correct problems.

Two weeks ago, the guys from Cisco and ALU showed up with things from and they can talk to each other, so we sent one of them, the Cisco guys lost, we sent them and the other guys got to go to Split and we were running 100 gig inter?operability between two vendors using our standard, and it worked for 24 hours without a single error. So you can plug vendor A into vendor B so the whole reason for having a transport system was now definitely gone.

So, we ?? everything just moves. If you have upper layers, that means the routing protocols, if the routing protocols knows the performance the optics, you don't need BFD. If the router layers knows that ?? with forward error correct you basically can get early warning, if you know that this circuit is about to break because we are going down a slippery slope, you can say increase the metric of this wire so traffic moves somewhere else before it even breaks, okay. And if you cut slowly with a pair of dikes you can actually make traffic move off the fibre before you cut it. So my by integrating layers we got better flexibility and performance and we can actually build an optimised network for moving packets and packets is the only thing we are moving. If it's not a packet it's like a single digit percent of your packet, stick it in a packet and it doesn't matter if you have a hundred percent. Make it look like a packet instead of trying to do it the other way around, what they are doing is building a network for the 5 percent traffic and putting everything else on top of it.

By not having filters, I can have any channel with anywhere in the network, I can use any encoding, any format I want as long as I make sure that I don't step on myself. Which means that I don't need to go out and change anything in the entire infrastructure for going, for a 400 gig or 1.6, it. The openly thing I need to do is change the two end points at point to point basis, so on a per link basis, I can change the two end points and move to out wear hardwaren coding and don't need to change anything except the two end points. That gives me better flexibility. And by not doing layer violation, by not having anything to lake from layer 1 and layer 2 when it comes to the data and information transfer, I really can have any kind of transport.

Now, how you run this network? Well, we wanted people to be up here. So we ended up saying okay, let's make service models. A service model is something describing what the network is doing in YANG. YANG is a language that we did in the IETF. You speak YANG to the network elements. You have the operator up here, at the end of the day you put the customer up here and I say here is what we want the network to do and we let it do it. Nobody talks to anything here.

And you go, is it possible? Yeah.

First drawing of how to to do this looked like this. It debenture work, we ended up looking what do we actually need. Here is the old legacy system. We said okay let's insert this below the legacy system. What you need to do is do order management work order, inventory, initiate diagnostics, you know, get paid, advance diagnostics, I don't know what that was the performance management, fault management and Fastrack order filters. Those were the interfaces we needed. So instead of this old system talking to the network, it talks to this box so if you can make an NNI, remember I said network to network interface on this box that talks to somebody items, somebody else can use your network resources the same way this guy can do.

The OSS gateway ends up looking something like this. You have the old systems. You convert the old system stuff into the NETCONF/YANG and we started off using a commercial product using NCS to hook this thing on to, and basically we built a message router and then you end up finding data store and we ended up building a distributed hash table that's where we store the table. All those traps actually goes through the database in order to avoid everybody talking to everything everywhere and we actually have all the data stored.

I tried to draw this picture differently. We had the YANG router and an event comes in: customer order missing feature, whatever, it goes ?? those different service models that may be software subscribes to the YANG router by saying here is a regular expression ?? this describes what kind of events I'm interested in. And you can say, do you want it informal? Or do you want to have it saying are you actually a decision making? So at the end of day, you can do capacity planning, service placement, all of this, all the scratch bade data is stored in the DHT. This is available to everybody. This describes how to find the data. The YANG models of the data describes how to find it in the database. At the end of day when everything passes you end up configuring network elements. We have two kind of routers in this network. I only know how to configure one of them so what I did I take the configure and I turned it to YANG and then I feed to this thing and it feeds to the router I don't understand. And I haven't logged in yet.

Here is in our problem. When every carry shows up and phone company shows up and they make a drawing of what they are doing, the customer is over here, mobile operators, they think they make more money the more boxes they have here. The customer is here and the Internet is here and they have all these boxes and they spend loads of money to get this done. My first question is okay, if everything is here, what's here? I go what? What we want to do is make the customer part of the Internet. That's why we are all here.

Here is the drawing of a typical setup today and this is taking the old model we have at IPv4 and junction changing IPv4 to IPv6, piece of cake. We could actually do this because the customer has a fair amount of address space, he has 250 free out of those if I used one for those link. He needs to get a block of addresses, he needs to allocate things. Okay. But you know what, it would probably look like this. And now we end up with another problem that I don't really know how to so. If this is the network, and we have policy, how does device sit inside the network tell which way to use? And here we have two ways of going, we can go that way or sneak on our neighbour's provider. This is interface inside our outside. That interface inside our outside? Are we going to have internal gateways or border gateways or are we just going to call them gateways and it doesn't matter in the topology where you are? And you basically have a way of distributing the policy among all your boxes or does it matter where the attacker is. In the attacker makes it through the firewall he is still being checked or not.

The real scenario is actually probably going to look like this. They have multiple service provider. He have a set top box sitting here. You want to play NetFlix. The kids want to play games on that way. All the way up to the player on the screen you probably need to be able to choose which server provider am I going to use? The way I thought put all the address blocks from all the service providers and we colour them inside the network. Here is a lot of works, look at this here. I used 239 networks to go, not in this picture, allocated to the customer. If grand ma is going to configure that we need to automate this and we need to find a way to make it done without using DHCP because I don't think ?? I don't configure my backbone with DHCP. Why would you configure this network as complicated as it is.

I think that we will use DHCP to give, hey, here is your address, here is your default gateway, you can't do that with DHCP in IPv6. Here is your NTP server. Here's your address. Why not have the home gateway to do exactly what we do with the big boxes. Use a YANG model, express what you want, send it off to the network that helps you configure your box, if you're grandma. If you are not grandma, you can do it all the way you want yourself.

I am done. Thank you so much.


JAN ZORZ: Peter, thank you very much for this presentation, it was a blast and you were in time with 45 minutes as we told you. And just as a note I was observing the room, I have seen no ?? I did not see that much attention in the room for any presentation up to date. It was really interesting. So, we have five minutes for questions and the discussion.

PETER LOTHBERG: First of all, every slide is a month of talks. There is a lot of things on this. I'm going to say the whole week, if you want to talk about anything on my talk, you want more documentation, catch me somewhere, I'm staying until Friday, and, that way, we can probably get some more ?? I want to have feedback on what I screwed up this time.

JAN ZORZ: So, where are the queues at the mics?

AUDIENCE SPEAKER: Peter, I have not one question but a minimum one. First of all, you showed us as a view of the world, very monolithic because in this model because 1, 2, data centre, in fact virtualised infrastructure normal to reach the netted infrastructure or or the netted infrastructure. But the ?? you have only once thing shows question with regard the connection because the only ?? need the connectivities and only one ISP. I can understand the imDeutsche Telecom. We have in Russia the same kind. You have some mix of business model behind this and technological toys, but for me, it's interesting what do you think personally, in this situation?


AUDIENCE SPEAKER: You draw us and to show us unanalytical infrastructure which can be useful for one provider, infrastructural one provider. It's not about Internet, it's about you ?? one of your first tests it was that your architecture based on the major proportion of traffic. You showed in your presentations it wasn't 50% it's Internet. But you indicated to one provider.

PETER LOTHBERG: So, you know, first of all, the first question I usually get is, how do you migrate from the old spaghetti to the new one? I said over build, let's go build a new network. Phase it in when all the equipment is dying. This picture shows three ISPs, I'm one, you are two and somebody else in this room is three. The home is multi?home and they are multi?choices. What's inside one is what I'm showing. What's inside two is your call. Am I supposed to build a network for everybody in my network? Then it's not my network. Then we should give it to the community.

AUDIENCE SPEAKER: This is a separate point. Are you ready for NSA being big or ?? in your home?

PETER LOTHBERG: That's they're problem. (Their) my job was, and I was very clear on that. My job was to try to come up with the best cost effective mix of how do we build infrastructure for the future? I have not tried to solve any business problem. I have just tried to minimise the cost, optimise the performance of the network and the flexibility. Now, if the business guys then tells me how to use this infrastructure, that's their call and not my department, right. I was ?? I just made sure that whatever crazy thing they came up with, I can do that. And if you want to do very, very complicated things with your packets, your service change things in the data centre, for example.

AUDIENCE SPEAKER: My first concern is because in fact your technical solution based on some business assumption.

PETER LOTHBERG: They are based on how many packets I need to deliver to the customer around I tried to do that as cheap as possible and the first thing you have to do is build something that doesn't have too much people talking.

JAN ZORZ: Can we go to the next question.

AUDIENCE SPEAKER: I'd like ?? HLL ?? I see it as a possibility for not monolithic network. In fact, for example, your business guys you open up your network for any Internet services on your network, it's just as good and while I thing you are presentation idea is moving from packets in different data technologies to services, which is pretty good thing, so there is no monolithic things that's involved...

JAN ZORZ: Was that a question or a statement?

AUDIENCE SPEAKER: It's a statement I'd like to pose, with his concern that your network is over?monolithic and there is no other items in there. In what way does he ?? and this is a question ?? do you see in your model that there is any other companies that operates EC which provides services, for example, BPN services on your network?

PETER LOTHBERG: So I think I understand the question. Let me answer what I think the question is. You can put ?? anybody can offer services on this network as long as they connect to the network and speak IPv6 packets.

AUDIENCE SPEAKER: And mark them.

PETER LOTHBERG: And if this network is reselling to somebody else, the game is that the interconnect is IPv6 for transport, the interconnect for configuring, monitoring and operating that network is the NNI between the OSS gateways, so if you are using this network to build MPLS, VPN something, but now you are into ?? sort of, I put the machinery there, and you can use my computer, you can put your own computer next to R2 but all of those are business things, it has nothing to do with architecture. It looks the same.

JAN ZORZ: We are running out of time here. So, thank you, Peter, for a great presentation.


JAN ZORZ: Now, we have one lightning talk, ten minutes. We have the speaker change at the last moment, Frank Le Gall will present the latest IPv6 survey results. Thank you.

SPEAKER: Hello everybody. As said, Frank Le Gall is doing the presentation but I do not make the work, so all this work was prepared by Maater Botterman, who is in the US today, so [I will] do the presentation today. What I will quickly show new ten minutes it's ?? about the survey which has been run, so the proposal of survey is to get an overview of what is the current situation regarding IPv6 deployment and what are the plans for the future. Looking especially within the require communities.

So this survey is prepared by Maarten Botterman in conjunction with the all five. It's a survey which is, let's say, short; I'll try to keep it short. And it has been run every year since nearly 2008 where ARIN made the first version of this survey which was then, in 2009, reused by RIPE NCC and APNIC, and then in 2010, doing the survey and all the RIRs are now involved in the survey.

This is about 50 slides tell me to keep it short. So, here is are few slides and at the end I will give you the link where you can get the full presentation of the survey.

So here are a few ?? the main question which we were asked. The first is about the percentage of customer bases making use of IPv6 connectivity and what we see is that, okay, with microscope, because it's still low, we are talking about numbers we have 1% of users but at least we see an increase over the years of the possibility of users make using IPv6. But still if you look at the top of the grid there is only 1% of the user so still very, very small.

Interesting thing is asking the ISPs are you interested in commenting a take of IPv6 to your customers and okay, produced here it was more or less a table but each year we see a slight increase with 20% of the ISPs now using the IPv6 to the customer.

Okay. The IPv6 presentation compared this year with last year we see more or less the same results. With 22% of ISPs having no IPv6 presence at all and the others having external or internal presence.

A question, you can see in the overall the difference between the one plan and the other one. The IPv6 the differences, on the encountered problems and they are quite different but asking the one that really made it, the first biggest problems they got was the lack of user demand and the technical problems, together with the question clearly ?? still expense sees missing on the IPv6 side.

A question which was introduced last year in the survey, it's a new question, so we do not have the retrospective view, but do ISP plan to use or make use of LSN in the lightning talk and still 85, 82% do not plan to make use of LSN but we can see a slow increase of ISPs that may be willing to make use of LSN.

And comparing the one we are using and the one we plan, comparing 2012 and 2013 for the one we are using that are using more in 2013, instead of IPv6, whereas for the one who plan to use it in 2013, less plan to use it instead of IPv6 but more along with IPv6.

Interesting things is when we asked people, okay, when do you plan, what is your timeline to implement IPv6? And we see more or less the same figure between this year and last year. The good news if you look at the figure this year, 70% of the people say okay, within one year they will deployed IPv6. So, next year, we can expect that 70% of the ISPs will have deployed IPv6. Unfortunately, it was what was said last year, so, to that note we can translate. At least we can see that people with are working on it, the same for offering IPv6 to the business customers. The situation is the same.

So, overall, most of of the ISPs offer IPv6 to their customers and more, up to 80% says that they will do so in the coming two years. We saw a slight increase in the usage of IPv6 and CGN is showing not used a replaced for IPv6. So that's not yet the situation.

So we there is still a community of people, having a lot of IPv4 addresses available but the organisers of the importance of moving to IPv6.

So, thank you to all who responded to the survey and those who would like to enter it again, probably the survey will take place again next year, so the one who didn't use it, will have the opportunity to do it next year. Thank you.


JAN ZORZ: Maybe we have time for one question... short...

So, no questions. So, we have a coffee break now. Thank you for being here and listening and see you in half an hour. Thank you.