Today's ARIN estimated depletion date:


Amount of IPv6 traffic

2011-11-24

There are plenty of numbers floating around about the amount of IPv6 traffic that can be seen. Whenever I go to an industry event, somebody is always presenting a number. All of them are probably accurate but they are all significantly different. What are they actually are measuring and why are they so different?
First let’s look at some numbers from different industry stakeholders:

Backbone
At the Texas IPv6 Summit in September 2011, Yves Poppe from TATA communication noted that 0.061% of his traffic was IPv6. That doesn’t seem to be a lot. To simplify things, let’s round that number to 0.1%
Slides can be found here (slide 9)

Content
At IETF81 in Quebec City in July a panel of Google, Facebook, Cisco and Yahoo! noted how much IPv6 traffic they could see on the world IPv6 day. slides can be found here.
Google could see 0.33%, facebook 0.20, Yahoo! 0.29% and Cisco could see 1.11%
This sounds better, but still not great. Let’s round this number of to 1%

Clients
Also at the TXv6TF Summit in Austin, Ron Broersma from DREN presented how much traffic he can see in his fully dual stacked network. An astonishing 10% of his traffic is over IPv6. That sounds like a lot. His slides can be found here

How come these numbers are so significantly different from each other? Well, it is because the measure different things. Remember that both a v6 capable source and an IPv6 capable destination is needed for an IPv6 session to take place. The number from the IPv6 capable content providers (1%) is measuring the amount of IPv6 capable clients. The number from DREN is actually measuring the number of IPv6 capable servers (10%). The number from the backbone provider is measuring the ratio of IPv6 traffic that is the mathematical product of the IPv6 capable clients and the IPv6 capable servers. The figure below illustrates what the different measurements actually measures.

Conclusion
The two conclusions that can be drawn from this exercise is that:
1. The number of IPv6 capable content far exceeds the number of IPv6 capable clients. This is alarming as it really points to the fact that service providers are not taking IPv6 seriously (there are a few exceptions).
2. Any additional clients that are being upgraded to support IPv6 will have larger impact on the total IPv6 that the Internet backbones can see than a similar increase for the IPv6 content. For example an additional 1% of clients would bring up the total Internet traffic to 0.2% (10% * 2%) whereas a 1% increase on the content side would only bring up the total backbone traffic to 0.11% (11% * 0.1%).

Tags: , , , , , ,

7 Responses to “Amount of IPv6 traffic”

  1. JMV2009 says on :

    Current ripe stock (in /8, including last /8):
    4.27 (ipv4depletion)/3.59(ripe)/3.51(ipv4report)/2.5 (tndh.net)

    After a lull in allocation rates, there has recently been an uptick in the recent ripe allocation rate which is more consistent with their long term usage.

    Ripe depletion rate is currently about 0.03 to 0.045 per week, depending on the range of dates you count a recent.

    Ripe depletion seems most likely in H2 2012, but last-minute acceleration can be a bith.

  2. Grahmd says on :

    Hi,
    Do maybe it’s 2 /8′s more, maybe 2 /8′s less, but depletion will still be in H2 2012? Did I got it right?

    Finally, can someone explain where the ipv4delpetion’s 4.27 comes from? I understand you can think it’s less than what’s shown at http://www.ripe.net/internet-coordination/ipv4-exhaustion/ipv4-available-pool-graph as it’s actually less if you look at http://albatross.ripe.net/delegated-extended/

    But how can you get to more than http://www.ripe.net/internet-coordination/ipv4-exhaustion/ipv4-available-pool-graph ?

  3. ipv4depletion says on :

    not sure where 4,27 came from. That is what Ripe had when Iana got depleted. it is less now. I need to fix my report so I can give you an exact number it broke down the other day.

    normally my number is a little less that Ripe’s official number because sometimes they do tests on blocks. those blocks are removed from the delegated file that i have to go by, but they are still considered free by Ripe.

  4. Grahm says on :

    Small note: according to http://www.ripe.net/internet-coordination/ipv4-exhaustion/ipv4-available-pool-graph

    this “little less that Ripe’s official number because sometimes they do tests on blocks”

    is by “8.45 million IPv4 addresses temporarily set aside for …”

  5. jmv2009 says on :

    Before the last-minute acceleration at the beginning of this year, apnic was running through addresses twice as fast ripe. Then APNIC started to accelerate at ~7 /8 left, and was empty within 3 months.

    RIPE only has ~2.5 (excluding last /8) left, so it could be empty within 3 months now, if it accelerates in the same way. It’s already a bit late! I think the three-month-worth-of-ip-addresses-at-a-time allocation policy is slowing down demand significantly. Apnic has a one year policy.

  6. Steve Hill says on :

    “The number of IPv6 capable content far exceeds the number of IPv6 capable clients. This is alarming as it really points to the fact that service providers are not taking IPv6 seriously”

    This doesn’t surprise me at all – at the moment, you have to go to some effort to implement IPv6, whether you are a content provider or a client. Content providers are generally a lot more technically capable and therefore can go to the effort of implementing IPv6. This is especially true of the big ones like Google, who actually have to forecast the future a bit instead of just reacting to the current situation, so actively know that they _need_ to implement v6 connectivity in the long term – for most of the small content providers, I suspect v6 isn’t even on their radar. As an example, I’ve been asking the datacentre that hosts some of my servers what their plans are for v6 for some time, and they have only recently started rolling it out (sadly not yet in the building where my servers are hosted) – this is an example of requiring effort; in order to roll out native v6 on my servers at the moment, I would have to physically move them to another datacentre (ok, I could use a tunnel broker, but yuck).

    On the other hand, a good chunk of the clients are home users and small (non-technical) businesses – the vast vast majority of them are nontechnical. They just want to plug their computer into the router and get internet access. I agree that the ISPs are largely not being helpful here, but even in cases where they are, you still need to go to some effort – EntaNet, for example, will give you native v6 over DSL, but you have to ask for it (i.e. you have to know that you want it, it doesn’t magically happen, so the vast majority of people won’t get it even though it is available). And even if the ISP did give you native v6 connectivity as standard, none of the consumer grade routers support it so again there is lots of effort involved. Until v6 “Just Works” for consumers, we’re going to contiue to see this shortfall, even when the ISPs have got their act together.

  7. George B says on :

    The amount of Internet traffic is no indication of the amount of v6 adoption out there. One problem is that we have places such as mobile device networks what are going out native v6 on their latest services/devices but people are still by and large trying to reach v4 services. This results in the traffic being converted to v4 before it leaves the user’s network. So it looks like just more v4 traffic.

    Until content providers of the popular services roll out native v6 content (without having to use some “special” name like “v6.example.com”) we aren’t going to know from traffic analysis how much v6 adoption there really is out there. Verizon’s new services are native v6 and so are other carriers around the world like Telefonica.

    You won’t have any idea how much v6 is out there until you start serving it. If everyone is waiting for the numbers to come up before they start serving content, everyone is going to be waiting forever.

Leave a Reply