Transition to IPv6: Have we reached?

The legendary story of the transition to IPv6 continued to surprise all of us. The RFC 2460 is the first complete IPv6 protocol specification. It was released in December 1998 and has been more than 20 years ago.

The whole significance of IPv6 is to specify the follow -up protocol of IPv4 because IPv4 address may be used up. However, we have used the IPV4 address more than ten years ago. The transition to IPv6 has been carried out for 20 years. If the prospect of the IPV4 address is injecting the sense of urgency, then we have lived in the end of 10 years. Therefore, it may be time to ask such a question: how long does this transition take?

This is a problem raised to a group at the recent Arin49 conference. It is foreseeable that there is no clear consensus for the answer. I want to discuss this issue in detail here.

A little history

In 1991, there was a problem with the Internet protocol (IP). At that time, it was still a small Internet, but the growth model was index level, and it would double every 12 months.

At that time, we had been emphasizing the IPv4 address pool of Class B. If there are no corrections, the address pool will be completely exhausted in 1994. We also put pressure on the routing system. The memory of the router deployed in 1992 only supports the growth of routing from the next 12 to 18 months. The comprehensive pressure of these routes and addressing was researched and discussed by the IETF Road working group (RFC 1380) at the time.

To solve this problem, there is a series of short -term, medium and long -term countermeasures.

In the short term, we abandoned the address plan based on class, but adopted a variable -size address prefix model. The routing protocol, including the Border Gateway Agreement (BGP), was quickly modified to support these types of unprepared addresses. However, the large -sized address prefix gives the address allocation process to increase the additional burden.

During the medium term, the regional Internet registration agency (RIP) enables each area to provide resources for its address allocation and registration functions; increase the particularity of address allocation and provide sufficient resources to allow more applications to apply relatively conservative address distribution strategies. And allow significant improvement of address efficiency. At the same time, the concept of "address sharing" using network address conversion (NAT) has also been supported in Internet service providers (ISP). NAT not only greatly simplifies the process of providing the service of ISP, but also plays an important role in reducing the pressure of address consumption.

Although these measures have postponed the two -year crisis to the managed ten -year -old scene, long -term technical response is required to expand the 32 -bit address field used in IPv4.

No such changes can be compatible with the installation foundation of the IPv4 system. Therefore, there are several different schools about what to do. One method is to jump and turn into a non -connected transmission configuration file using the OSI protocol kit, and use the OSI NSAP address in this process. The other is to change the IP as little as possible except the size of the address field. People have also put forward several other ideas on major changes to the IP protocol model.

Until 1994, IETF had made a method to determine the minimum change, namely IPv6. The address field was extended to 128 -bit, the flow ID field was introduced, the shard behavior was changed and introduced a optional header, and the address parsing protocol (ARP) was replaced by multiple broadcasts.

The bottom line is a new feature that IPV6 does not provide any IPv4. It has not introduced any major changes to the operation of the IP protocol. IPv6 is just an IP protocol with larger addresses.

transition

Although the design of IPv6 has spent a lot of energy, the concept of transitioning the network from IPv4 to IPv6 has not received much attention.

In view of the widespread adoption of IPv4, a naive expectation is that IPv6 will take off similarly, and there is no need to consider so much. In the first stage, we expect to see applications, hosts and networks, in addition to IPv4, also increase support for IPv6, and transform the Internet into a dual -stack environment. In the second stage, we can gradually cancel support for IPv4.

There are several problems in this plan! Perhaps the worst is the problem of resource allocation. The development of the Internet is very fast, and most of our efforts are committed to keeping up with demand. More users, larger capacity, more powerful servers, more content, more convenient services, more secure and better defense systems -all these revolves around a common theme -scale. Therefore, we can concentrate resources to meet the expanded needs, or are committed to the deployment of IPv6.

The short -term and medium -term measures we have taken have solved the urgent problem of depletion of the address. Therefore, for the industry, the scale expansion has a higher priority than the transition to IPv6.

In the mid -2000s, with the launch of the iPhone and its derivative products, the scale expansion accelerated to a brand new order. Suddenly, this is not just the scale of family and enterprises, but the scale of individual and liquidity.

IPv6 has become a necessary reasons formed, but we are not ready to deploy IPv6 as a response. Therefore, we added the consumption of the remaining IPv4 address and supported the first wave of large -scale mobile services with IPV4. The dual stack was not even an option in the mobile field at that time. The economics of financing for 3G infrastructure is quite strange, which means that the establishment of dual -stack infrastructure on the 3G platform is unrealistic. At the same time, the nature of the Internet decentralization hinders the efforts to transition to IPv6. If there is no host to integrate IPv6 into its network protocol stack, what is the significance of developing applications that support IPv6? If IPv6 support does not provide IPv6, what is the meaning of adding IPv6 to the host network protocol stack? If there is no host and application using IPv6, what's the point of IPv6 in IPv6? So nothing happened.

First of all, the various operating systems that are trying to break this kind of dependence are all kinds of operating systems. The full -featured IPv6 protocol stack is added to various versions of Linux, Windows, and Mac OS, and the mobile host protocol stack of iOS and Android.

But even so, it is not enough to make the transitional progress. It can be said that this situation has made the situation of IPv6 worse and delayed the transition for several years. The problem is that for hosting IPv6, people want to use IPv6. However, these hosts are IPv6 islands located in the ocean of IPv4.

The efforts of transition are concentrated on various tunnel methods for tunneling IPv6 packets through IPv4 networks. Although this operation can be manually executed when controlling the endpoints of the two tunnels, this is not a useful method. What we want is an automatic tunnel mechanism that can take care of all these details.

The first method is 6TO4. The first problem of 6TO4 is that it requires the public IPv4 address, so it cannot provide services to the IPv6 host behind NAT. The more critical issue is that the firewall does not know how to deal with the 6TO4 packet. When there is doubt, the default operation is to refuse to visit. Therefore, 6TO4 leads to a 20%failure rate and cannot be used. The NAT problem is also a problem, so people have designed the second automatic tunnel mechanism to perform NAT perception and penetration. As far as the failure rate is concerned, this mechanism is even worse -about 40%of the Teredo connection attempts to be observed is failed.

These initial transition tools are not only very poor, because they are very unreliable. Even when they work, the connection is very fragile, even slower than IPv4. The result is foreseeable. Not only the transition mechanism, but even the IPv6 itself has caused criticism.

Until 2011, IPv6 was largely ignored. Only a few service providers try to deploy IPv6. However, in each attempt, these service providers found that themselves and their suppliers must solve a series of special challenges. If there is no rich content and services on IPv6, the value of the entire practice is very low! So nothing happened.

Finally, there is action!

Until early 2011, the central IPv4 address pool managed by IANA was exhausted. The first RIR general distribution pool in April that year was exhausted, which attracted more attention from the ISP industry.

At first, the transition to IPv6 was faltering, but in the past ten years, the transition has accumulated motivation. Figure 1 shows the estimated share of IPv6 users who can access the IPv6 service.

Figure 1 The usage of IPv6 among all Internet users -from 2012 to the present

Source: APNIC Laboratory

From 5%in early 2016 to 17%at the end of 2017, the number of IPv6s increased by more than double. This is largely due to the rapid deployment of IPv6 in the Indian mobile network.

In the following three years (2018 ~ 2020), this number rose from 17%to 30%. The second growth stage is largely due to the deployment of IPv6 in China.

However, this is not just a story about deployment in India and China. The IPv6 capabilities of other economies have also continued to grow steadily in the past five years, such as Mexico, Brazil and the United States. If we take the deployment ratio of IPv6 as a measure of the user -about 30%of all users -then we have reasons to draw conclusions and the transition is going smoothly. Obviously, these 30%are not evenly distributed. When we consider the national economy, we will also see different pictures of this transition in the current state (Figure 2).

Figure 2 deployment of IPv6 in April 2022

Source: APNIC Laboratory

The transition to IPv6 is concentrated in a relatively small economy. The global average of only 32 economies of IPv6 is higher than 30%. From a regional point of view, the usage rate of IPv6 seems to be the highest in South Asia, North America and Western Europe, while in Africa and Pacific (Oceania) the lowest.

However, this is more likely to be misleading at the economy level, because they (1.4 billion people) and Pitkine (50 people) are equally positioned. If we observe the 10 largest IPv6 user group, we will get a slightly different view (Table 1). Table 1 IPv6 user quantity

Source: APNIC Laboratory

This data confirmed that in many parts of the world, the deployment of IPv6 in the large -scale consumer network has made progress smoothly.

how much longer?

Now that we are in the process of transition, the next question is how long will this transition last?

This seems to be a simple problem, but it does need more explanations. When we can announce the end of the transition, what is the "end point"? Or, when will this transition "complete"?

When there is no more traffic based on IPv4 on the Internet? Or is there no need for IPv4 on the Internet? Or when only IPv6 services are provided?

Maybe we should look at the market of IPv4 address. When the price of IPv4 address is completely collapsed, this transition end is defined?

Maybe we can stand on a pragmatic stand, not to complete the transitional definition when the Internet will no longer use the IPv4 address at all, but will not need to use the IPV4 address when the transition definition is not needed. This means that when the service provider uses IPv6 and does not support the IPv4 access mechanism to run an Internet service, we complete this transition.

what does that mean? Of course, ISPS needs to provide IPv6. But at the same time, all connected edge networks and hosts in these networks need to support IPv6. After all, ISP does not have an IPV4 service when completing the transition. This also means that all services used by the ISP client must be accessed through IPv6. Yes, this includes all popular cloud services and cloud platforms, all content stream media and all content distribution platforms. It also includes special platforms, such as Slack, Xero, Atlassian and similar platforms.

We must not reach this today, and it is unlikely to be achieved in the next two or three years. According to the data released by the World IPv6 release website, only about 30%of the websites in the first 1,000 websites can be visited through IPv6. Obviously, various service platforms have jobs to do.

In the discussions of the Arin 49 group, there are various views such as at least ten years to a quarter of the transition period.

Why does this transition take so long?

Obviously, the reason why the transition is so long is because there is no sense of urgency. Although some actors have made necessary changes to deploy dual -stack services and infrastructure, other actors do not think such actions are current priorities. As long as any such action will not have unacceptable business consequences, delay will continue.

Part of the reason is that IPv6 has not obtained the main competitive advantage. It does not provide any special features of more efficiency, lower cost or unique service configuration files than IPv4. The initial stage of running the dual protocol stack environment means higher costs, and the benefits are very small.

The initial reason to use IPv6 was based on risk avoidance. NAT -form address sharing is considered an unacceptable measure. When the available IPv4 address pool is exhausted, it is considered that the further growth of the Internet will be impossible because we will fully abandon the use of NAT. Adopting IPv6 in time is considered a necessary measure to avoid this situation.

However, NAT was not an ISP problem at first. ISP effectively pack NAT outsourcing to the user terminal device (CPE) supplier, thereby pushing the entire problem to end users and applications. Application designers are facing a simple reality, either their applications run perfectly under the presence of NAT, or they cannot work. The application no longer uses the end -to -end connection model, but uses the server/client service model.

Next is a mobile network deployed on the connected client. ISP has deployed NAT to its own infrastructure, and the so -called operator -level NAT (CGN) has become commonplace. If we complete the transition before the large -scale popularization of mobile services, all these are avoided. However, because this occurs during the transition, each mobile service provider must launch a solution to provide IPv4 services to its customer base, whether they have deployed IPv6. This or that form of CGN is a compulsory requirement for everyone. IPv6 is an extra optional content.

Obviously, in different parts of the Internet, people feel the pressure of starting dual -stack services to varying degrees. Some operators have to urgently deploy IPv6 to reduce the pressure of insufficient public IPv4 address and support their CGN. Other operators do not have any deployment pressure at present, they are willing to wait for the right time to deploy.

Send a signal of common needs

Remember that the Internet is not a single entity. There are many network compositions, including consumer retail services networks, corporate networks, IoT (IoT), etc.

There are various suppliers in the Internet ecosystem. There are IP access operators, IP transmission providers, platform providers, chip manufacturers, application providers, content platforms, etc. These individual actors usually communicate their own needs through market signals, and market signals usually show prices. Commodities and services under demand are under price pressure, and the price and services that are no longer needed to withstand the price plunge. However, for many years, we have not allowed this market -based distribution mechanism to affect the IP address. There are many reasons behind this, including considering that address is a public item, hoping to implement the principle of protection of the address of the address to achieve a certain form of address obtaining fairness and prevent the distribution of address distribution due to market distortion.

As a result, the traditional pricing mechanism cannot issue the upcoming IPv4 scarce signal. Only when the market transfer market was added to the major environment in the 2010s, pricing information was effective as a market signal.

Over time, the price record of the transferred IPv4 address is shown in Figure 3.

Figure 3 The price of IPv4 address from 2014 ~ 2022

Source: Hilco StreamBank

If price fluctuations are a signal of market emotional changes, then from the beginning of 2014 to early 2018, public emotions have no substantial changes. A relatively slow price increase indicates that there is no concern for the shortage of IPv4 address supply that will appear. Even if prices rose in 2018, it was offset by stable prices in 2019 and 2020.

In 2021, the situation changed fundamentally, and the price of the IPv4 address doubled. The continuous rise in prices indicates that the market is in short supply.

The market is not good at predicting the future, but it helps us understand the current stage.

What does the current address price rising tell us? This seems to be saying that the transition is about to complete this incident without a consensus, and the demand for IPv4 address will continue to exceed the supply for a period of time. But as for how long this situation will last, market pricing information does not show.

When the transition to IPv6 is about to end, what do we want to see? It is speculated that by then, there will be no further demand for IPv4 -the market position of IPV4 will shift from demand to overdoing. In this case, the transition is about to end, and there will be no prospects for recovery. This will cause market prices to plummet.

Does the transition take a long time?

If we can't predict when this transition is over, can we at least at least do this transition form an opinion?

What shocked me was that human beings tend to treat the current situation as some form of impose arrears. Now we have been in the IPv6 transition for about 25 years. The Internet seems to be very practical for users and their services. Then we naturally believe that this situation will continue for a while, or even endlessly.

However, you can draw such a conclusion that whether this situation can stand on your feet depends on the Internet growth model you are ready to believe. If the growth stage of the Internet is actually over, what we see now is a saturated market, so yes, we are now in a state of maintaining. However, in the current situation, considering that the newly joined person still needs some form of IPv4 address access, and as the growth pressure continues, this situation will become increasingly difficult to maintain. This means that the establishment of a market that has established, including mass market housing services in many parts of the world, does not necessarily face any major pressure on expanding its platforms and obtaining more IPv4 addresses. But this is by no means all the network environment.

The area of ​​growth seems to be the continuous expansion of the field of cloud services and the Internet of Things, as well as the traditional retail Internet access market in Africa and parts of Asia. At present, it is estimated that among the 7.8 billion people in the world, the population of the Internet users is about 4.2 billion. There are still some room for growth in the Internet.

Is it the only choice to transition to IPv6?

The discussion about transitions assumes that IPv6 is the only available option. If this is the case, the scope of the digital system continues to expand to the world of embedded objects, which will make the pure IPv6 service inevitably sooner or later. Maybe we should ask whether this group of very limited options, that is, whether a choice is accurate.

Over the years, we have explored several different network models, one of which seems to provide potential feasible alternatives, namely name-based networks. In the initial concept of the name of the name, we consider using the service identifier name to replace the address field in the packet header, and based on these name routing and forwarding the data packet. When observing a large enough distance, the endpoint logo based on integer assignment and the endpoint identification based on a string -based coding form is largely homogeneous. The key is that the name of the name is not a special concept. The essential difference is that there is a larger token space with a sparse use mode.

However, what we built in the fierce competition of IPv4 is a bit different. When we analyze the service name in DNS, DNS tries to provide us with an address that allows access to the service. The address is not necessarily the only name of the specific service, because the analysis query of different DNS service names may also return the same IP address. The address is not fixed in time, because the same DNS parsing query can cause different addresses at different times. And the address is not common, because different users perform the same DNS parsing query that may get different addresses. The potential ambiguity in the service name to the IP address mapping can be solved at the transmission and application level. When connected to the address, the client needs to name the service it wants to connect. This explicit service logo is part of the client request exchange in the TLS protocol, part of the initial exchange of HTTP, and part of many other service protocols. This indirect level aims to allow one server to carry multiple services and allow one service to be carried on multiple servers. As a result, the IP address is not unique to the service or network terminal. Today's network is a version of the naming network. The role of the IP address is a short -term session -level token. It allows the network to distinguish between concurrent data packages and anything else.

When we continue to explore this digital space, suppose we design the structure of the group exchange network in the 1970s as the only architecture that can be worked, and we will never propose different architectures, which is a bit reversed. Today's NAT, Broadcasting System, content distribution network, and increasing application -level functions are pointing to another change in the Internet. It no longer depends on the IP layer of the general -purpose adaptation protocol between different network media and applications, but increases the co -dependent protocol stack to the application layer.

Have we reached ipv6?

Obviously not yet.

But we are approaching. If the end point is to provide customers with pure IPv6 services, the continuous efforts of using IPv6 in the field of content and service platform will have some impact on when we reach this point.

I did not see the public data about the number of IPv4 addresses used in CGN and the supply rules of the user scale. However, with the increase in IPv6 in the service platform and the use of IPv6 preference rules in the application, I definitely believe that the proportion of the size of the IPv4 address pool to the customer base will gradually decline over time.

This leads to the second point: we are likely not to reach the end of this transition vigorously, but to arrive silently. Not all ISP will turn off their CGN on a specific date and exit IPv4. A more likely situation is that the size of the IPV4 address pool required for the ISP client will continue to decline. To some extent, it has no commercial value for ISP's own infrastructure to continue to operate these CGN services. Customers who need IPv4 services need to find an ISP that still provides IPv4, or switch to external IPv4 services, and use a form of VPN tunnel to access it. This may not happen at the same time at the same time, but occur in a long period of time in a long period of time.

So, when will this transition end?

I still don't know!

Author: APNIC chief scientist Geoff Huston

Source: APNIC

Translation: Dong Ming, Northeast University

- END -