| IPv6 Operations | J. Livingood |
| Internet-Draft | Comcast |
| Intended status: Informational | February 21, 2012 |
| Expires: August 22, 2012 |
Considerations for Transitioning Content to IPv6
draft-ietf-v6ops-v6-aaaa-whitelisting-implications-09
This document describes considerations for the transition of end user content on the Internet to IPv6. While this is tailored to address end user content, which is typically web-based, many aspects of this document may be more broadly applicable to the transition to IPv6 of other applications and services. This document explores the challenges involved in the transition to IPv6, potential migration tactics, possible migration phases, and other considerations. The audience for this document is the Internet community generally, particularly IPv6 implementers.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on August 22, 2012.
Copyright (c) 2012 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
This document describes considerations for the transition of end user content on the Internet to IPv6. While this is tailored to address end user content, which is typically web-based, many aspects of this document may be more broadly applicable to the transition to IPv6 of other applications and services. The issues explored herein will be of particular interest to major web content sites (sometimes described hereinafter as "high-service-level domains"), which have specific and unique concerns relating to maintaining a high-quality user experience for all of their users during their transition to IPv6. This document explores the challenges involved in the transition to IPv6, potential migration tactics, possible migration phases, and other considerations. Some sections of this document also include information about the potential implications of various migration tactics or phased approaches to the transition to IPv6.
The goal in transitioning content to IPv6 is to make that content natively dual-stack enabled, which provides native access to all end users via both IPv4 and IPv6. However, there are technical and operational challenges in being able to transition smoothly for all end users, which has led to the development of a variety of migration tactics. A first step in understanding various migration tactics is to first outline the challenges involved in moving content to IPv6.
Implementers of these various migration tactics are attempting to protect users of their services from having a negative experience (poor performance) when they receive DNS responses containing AAAA resource records or when attempting to use IPv6 transport. There are two main concerns which pertain to this practice; one of which is IPv6-related impairment and the other which is the maturity or stability of IPv6 transport (and associated network operations) for high-service-level domains. Both can negatively affect the experience of end users.
Not all domains may face the same challenges in transitioning content to IPv6, since the user base of each domain, traffic sources, traffic volumes, and other factors obviously will vary between domains. As a result, while some domains have used an IPv6 migration tactic, others have run brief IPv6 experiments and then decided to simply turn on IPv6 for the domain without further delay and without using any specialized IPv6 migration tactics [Heise]. Each domain should therefore consider its specific situation when formulating a plan to move to IPv6; there is not one approach that will work for every domain.
Some implementers have observed that when they added AAAA resource records to their authoritative DNS servers in order to support IPv6 access to their content that a small fraction of end users had slow or otherwise impaired access to a given web site with both AAAA and A resource records. The fraction of users with such impaired access has been estimated to be as high as 0.078% of total Internet users [IETF-77-DNSOP] [NW-Article-DNSOP] [IPv6-Growth] [IPv6-Brokenness].
While it is outside the scope of this document to explore the various reasons why a particular user's system (host) may have impaired IPv6 access, and the potential solutions [I-D.ietf-v6ops-happy-eyeballs] [RFC6343], for the users who experience this impairment it has a very real performance impact. It would impact access to all or most dual stack services to which the user attempts to connect. This negative end user experience can range from somewhat slower than usual access (as compared to native IPv4-based access), to extremely slow access, to no access to the domain's resources whatsoever. In essence, whether the end user even has an IPv6 address or not, merely by receiving a AAAA record response the user either cannot access a Fully Qualified Domain Name (FQDN, representing a service or resource sought) or it is so slow that the user gives up and assumes the destination is unreachable.
Some implementers have discovered that network operations, operations support and business support systems, and other operational processes and procedures are less mature for IPv6 as compared to IPv4, since IPv6 has not heretofore been pervasively deployed. This operational immaturity may be observed not just within the network of a given domain but also in any directly or indirectly interconnected networks. As a result, many domains consider it prudent to undertake any network changes which will cause traffic to shift to IPv6 gradually in order to provide time and experience for IPv6 operations and network practices mature.
While Section 2.2 pertains to risks due to immaturity in operations, a related concern is that some technical issues may not become apparent until some moderate to high volume of traffic is sent via IPv6 to and from a domain. As above, this may be the case not just within the network of that domain but also for any directly or indirectly interconnected networks. Furthermore, compared to domains with small to moderate traffic volumes, whether by the count of end users or count of bytes transferred, high-traffic domains receive such a level of usage that it is prudent to undertake any network changes gradually and in a manner which minimizes the risk of disruption. One can imagine that for one of the top ten sites globally, for example, the idea of suddenly turning on a significant amount of IPv6 traffic is quite daunting and would carry a relatively high risk of network and/or other disruptions.
It is important that the challenges in transitioning content to IPv6 noted in Section 2 are addressed, especially for high-service-level domains. Some high-service-level domains may find the prospect of transitioning to IPv6 extremely daunting without having some ability to address these challenges and to incrementally control their transition to IPv6. Lacking such controls, some domains may choose to substantially delay their transition to IPv6. A substantial delay in content moving to IPv6 could certainly mean there are somewhat fewer motivating factors for network operators to deploy IPv6 to end user hosts (though they have many significant motivating factors that are largely independent of content). At the same time, unless network operators transition to IPv6, there are of course fewer motivations for domain owners to transition content to IPv6. Without progress in each part of the Internet ecosystem, networks and/or content sites may delay, postpone, or cease adoption of IPv6, or to actively seek alternatives to it. Such alternatives may include the use of multi-layer or large scale network address translation (NAT), which is not preferred relative to native dual stack.
Obviously, transitioning content to IPv6 is important to IPv6 adoption overall. While challenges do exist, such a transition is not an impossible task for a domain to undertake. A range of potential migration tactics, as noted below in Section 4, can help meet these challenges and enable a domain to successfully transition content and other services to IPv6.
Domains have a wide range of potential tactics at their disposal that may be used to facilitate the migration to IPv6. This section includes many of the key tactics that could be used by a domain but it is by no means an exhaustive or exclusive list. Only a specific domain can judge whether or not a given (or any) migration tactic applies to their domain and meets their needs. A domain may also decide to pursue several of these tactics in parallel. Thus, the usefulness of each tactic and the associated pros and cons will vary from domain to domain.
Domains can endeavor to fix the underlying technical problems experienced by their end users during the transition to IPv6, as noted in Section 2.1. One challenge with this option is that a domain may have little or no control over the network connectivity, operating system, client software (such as a web browser), and/or other capabilities of the end users of that domain. In most cases a domain is only in a position to influence and guide their end users. While this is not the same sort of direct control which may exist in an enterprise network for example, major domains are likely to be trusted by their end users and may therefore be able to influence and guide these users in solving any IPv6-related impairments.
Another challenge is that end user impairments are something that one domain on their own cannot solve. This means that domains may find it more effective to coordinate with many others in the Internet community to solve what is really a collective problem that affects the entire Internet. Of course, it can sometimes be difficult to motivate members of the Internet community to work collectively towards such a goal, sharing the labor, time, and costs related to such an effort. However, World IPv6 Day [W6D] shows that such community efforts are possible and despite any potential challenges, the Internet community continues to work together in order to solve end user IPv6 impairments.
One potential tactic may be to identify which users have such impairments and then to communicate this information to affected users. Such end user communication is likely to be most helpful if the end user is not only alerted to a potential problem but is given careful and detailed advice on how to resolve this on their own, or is guided to where they can seek help in doing so. Another potential tactic is for a domain to collect, track over time, and periodically share with the Internet community the rate of impairment observed for a domain. In any such end user IPv6-related analysis and communication, Section 6.2 is worth taking into account.
However, while these tactics can help reduce IPv6-related impairments Section 2.1, they do not address either operational maturity concerns noted in Section 2.2 or volume-based concerns noted in Section 2.3, which should be considered and addressed separately.
Another potential migration tactic is for a domain to gain experience using a special Fully-Qualified Domain Name (FQDN). This has become typical for domains beginning the transition to IPv6, whereby an address-family-specific name such as ipv6.example.com or www.ipv6.example.com is used. An end user would have to know to use this special IPv6-specific name; it is not the same name used for regular traffic.
This special IPv6-specific name directs traffic to a host or hosts which have been enabled for native IPv6 access. In some cases this name may point to hosts which are separate from those used for IPv4 traffic (via www.example.com), while in other cases it may point to the same hosts used for IPv4 traffic. A subsequent phase, if separate hosts are used to support special IPv6-specific names, is to move to the same hosts used for regular traffic in order to utilize and exercise production infrastructure more fully. Regardless of whether or not dedicated hosts are used, the use of the special name is a way to incrementally control traffic as a tool for a domain to gain IPv6 experience and increase IPv6 use on a relatively controlled basis. Any lessons learned can then inform plans for a full transition to IPv6. This also provides an opportunity for a domain to develop any necessary training for staff, to develop IPv6-related testing procedures for their production network and lab, to deploy IPv6 functionality into their production network, and to develop and deploy IPv6-related network and service monitors. It is also an opportunity to add a relatively small amount of IPv6 traffic to ensure that network gear, network interconnects, and IPv6 routing in general is working as expected.
While using a special IPv6-specific name is a good initial step to functionally test and prepare a domain for IPv6, including developing and maturing IPv6 operations, as noted in Section 2.2, the utility of the tactic is limited since users must know the IPv6-specific name, the traffic volume will be low, and the traffic is unlikely to be representative of the general population of end users (they are likely to be self-selecting early adopters and more technically advanced than average), among other reasons. As a result, any concerns and risks related to traffic volume as noted Section 2.3 should still be considered and addressed separately.
Another potential tactic, especially when a high-service-level domain is ready to move beyond an IPv6-specific name, as described in Section 4.2, is to selectively return AAAA resource records (RRs), which contain IPv6 addresses. This selective response of DNS records is performed by an authoritative DNS servers for a domain in response to DNS queries sent by DNS recursive resolvers [RFC1035]. This is commonly referred to in the Internet community as "DNS Resolver Whitelisting", and will be referred to as such hereafter, though in essence it is simply a tactic enabling the selective return of DNS records based upon various technical factors. An end user is seeking a resource by name, and this selective response mechanism enables what is perceived to be the most reliable and best performing IP address family to be used (IPv4 or IPv6). It shares similarities with Content Delivery Networks, Global Server Load Balancing, DNS Load Balancing, and Split DNS, as described below in Section 4.3.2, Section 4.3.3, Section 4.3.4. A few high-service-level domains have either implemented DNS Resolver Whitelisting (one of many migration tactics they have used or are using) or are considering doing so [NW-Article-DNS-WL] [WL-Ops].
This is a migration tactic used by domains as a method for incrementally transitioning inbound traffic to a domain to IPv6. If an incremental tactic like this is not used, a domain might return AAAA resource records to any relevant DNS query, meaning the domain could go quickly from no IPv6 traffic to potentially a significant amount as soon as the AAAA resource records are published. When DNS Resolver Whitelisting is implemented, a domain's authoritative DNS will selectively return a AAAA resource record to DNS recursive resolvers on a whitelist maintained by the domain, while returning no AAAA resource records to DNS recursive resolvers which are not on that whitelist. This tactic will not have a direct impact on reducing IPv6-related impairments Section 2.1, though it can help a domain address operational maturity concerns Section 2.2 and concerns and risks related to traffic volume Section 2.3. While DNS Resolver Whitelisting does not solve IPv6-related impairments, it can help a domain to avoid users that have them. As a result, the tactic removes their impact in all but the few networks that are whitelisted. DNS Resolver Whitelisting also allows a website operator to protect non-IPv6 networks (i.e. networks that do not support IPv6 and/or do not have plans to do so in the future) from IPv6-related impairments in their networks. Finally, domains using this tactic should understand that the onus is on them to ensure that the servers being whitelisted represent a network that has proven to their satisfaction that they are IPv6-ready and this will not create a poor end user experience for users of the whitelisted server.
There are of course challenges and concerns relating to DNS Resolver Whitelisting. Some of the concerns with a whitelist of DNS recursive resolvers may be held by parties other than the implementing domain, such as network operators or end users that may not have had their DNS recursive resolvers added to a whitelist. Additionally, the IP address of a DNS recursive resolver is not a precise indicator of the IPv6 preparedness, or lack of IPv6-related impairment, of end user hosts which query (use) a particular DNS recursive resolver. While the IP addresses of DNS recursive resolvers on networks known to have deployed IPv6 may be an imperfect proxy for judging IPv6 preparedness, or lack of IPv6-related impairment, it is one of the better available methods at the current time. For example, implementers have found that it is possible to measure the level of IPv6 preparedness of the end users behind any given DNS recursive resolver by conducting ongoing measurement of the IPv6 preparedness of end users querying for one-time-use hostnames and then correlating the domain's authoritative DNS server logs with their web server logs. This can help implementers form a good picture of which DNS recursive resolvers have working IPv6 users behind them and which do not, what the latency impact of turning on IPv6 for any given DNS recursive resolver is, etc. In addition, given the current state of global IPv6 deployment, this migration tactic allows content providers to selectively expose the availability of their IPv6 services. While opinions in the Internet community concerning DNS Resolver Whitelisting are understandably quite varied, there is clear consensus that DNS Resolver Whitelisting can be a useful tactic for use during the transition of a domain to IPv6. In particular, some high-service-level domains view DNS Resolver Whitelisting as one of the few practical and low-risk approaches enabling them to transition to IPv6, without which their transition may not take place for some time. However, there is also consensus that this practice is workable on a manual basis (see below) only in the short-term and that it will not scale over the long-term. Thus, some domains may find DNS Resolver Whitelisting a beneficial temporary tactic in their transition to IPv6.
At the current time, generally speaking, a domain that implements DNS Resolver Whitelisting does so manually. This means that a domain manually maintains a list of networks that are permitted to receive IPv6 records (via their DNS resolver IP addresses) and that these networks typically submit applications, or follow some other process established by the domain, in order to be added to the DNS Whitelist. However, implementers foresee that a subsequent phase of DNS Resolver Whitelisting is likely to emerge in the future, possibly in the near future. In this new phase a domain would return IPv6 and/or IPv4 records dynamically based on automatically detected technical capabilities, location, or other factors. It would then function much like (or indeed as part of) global server load balancing, a common practice already in use today, as described in Section 4.3.2. Furthermore, in this future phase, networks would be added to and removed from a DNS Whitelist automatically, and possibly on a near-real-time basis. This means, crucially, that networks would no longer need to apply to be added to a whitelist, which may alleviate many of the key concerns that network operators may have with this tactic when it is implemented on a manual basis.
Using a "whitelist" in a generic sense means that no traffic (or traffic of a certain type) is permitted to the destination host unless the originating host's IP address is contained in the whitelist. In contrast, using a "blacklist" means that all traffic is permitted to the destination host unless the originating host's IP address is contained in the blacklist. In the case of DNS Resolver Whitelisting, the resource that an end user seeks is a name, not an IP address or IP address family. Thus, an end user is seeking a name such as www.example.com, without regard to the underlying IP address family (IPv4 or IPv6) which may be used to access that resource.
DNS Resolver Whitelisting is implemented in authoritative DNS servers, not in DNS recursive resolvers. These authoritative DNS servers selectively return AAAA resource records using the IP address of the DNS recursive resolver that has sent it a query. Thus, for a given operator of a website, such as www.example.com, the domain operator implements whitelisting on the authoritative DNS servers for the domain example.com. The whitelist is populated with the IPv4 and/or IPv6 addresses or prefix ranges of DNS recursive resolvers on the Internet, which have been authorized to receive AAAA resource record responses. These DNS recursive resolvers are operated by third parties, such as Internet Service Providers (ISPs), universities, governments, businesses, and individual end users. If a DNS recursive resolver is not matched in the whitelist, then AAAA resource records WILL NOT be sent in response to a query for a hostname in the example.com domain (and an A record would be sent). However, if a DNS recursive resolver is matched in the whitelist, then AAAA resource records WILL be sent. As a result, while Section 2.2 of [RFC4213] notes that a stub resolver can make a choice between whether to use a AAAA record or A record response, with DNS Resolver Whitelisting the authoritative DNS server can also decide whether to return a AAAA record, an A record, or both record types.
When implemented on a manual basis, DNS Resolver Whitelisting generally means that a very small fraction of the DNS recursive resolvers on the Internet (those in the whitelist) will receive AAAA responses. The large majority of DNS recursive resolvers on the Internet will therefore receive only A resource records containing IPv4 addresses. When implemented manually, domains may find the practice imposes some incremental operational burdens insofar as it can consume staff time to maintain a whitelist (such as additions and deletions to the list), respond to and review applications to be added to a whitelist, maintain good performance levels on authoritative DNS servers as the whitelist grows, create new network monitors to check the health of a whitelist function, perform new types of troubleshooting related to whitelisting, etc. In addition, manually-based whitelisting imposes some incremental burdens on operators of DNS recursive resolvers (such as network operators), since they will need to apply to be whitelisted with any implementing domains, and will subsequently need processes and systems to track the status of whitelisting applications, respond to requests for additional information pertaining to these applications, and track any de-whitelisting actions.
When implemented on an automated basis in the future, DNS recursive resolvers listed in the whitelist could expand and contract dynamically, and possibly in near-real-time, based on a wide range of factors. As a result, it is likely that the number of DNS recursive resolvers on the whitelist will be substantially larger than when such a list is maintained manually, and it is likely the the whitelist will grow at a rapid rate. This automation can eliminate most of the significant incremental operational burdens on both implementing domains as well as operators of DNS recursive resolvers, which is clearly a factor that is motivating implementers to work to automate this function.
Section 4.3.1.1 and Figure 1 have more details on DNS Resolver Whitelisting generally. In addition, the potential deployment models of DNS Resolver Whitelisting (manual and automated) are described in Section 5. It is also important to note that DNS Resolver Whitelisting also works independently of whether an authoritative DNS server, DNS recursive resolver, or end user host uses IPv4 transport, IPv6, or both. So, for example, whitelisting may not result in the return of AAAA responses even in those cases where the DNS recursive resolver has queried the authoritative server over IPv6 transport. This may also be the case in some situations when the end user host's original query to its DNS recursive resolver was over IPv6 transport, if that DNS recursive resolver is not on a given whitelist. One important reason for this is that even though the DNS recursive resolver may have no IPv6-related impairments, this is not a reliable predictor of whether the same is true of the end user host. This also means that a DNS whitelist can contain both IPv4 and IPv6 addresses.
Specific implementations will vary from domain to domain, based on a range of factors such as the technical capabilities of a given domain. As such, any examples listed herein should be considered general examples and are not intended to be exhaustive.
The system logic of DNS Resolver Whitelisting is as follows:
+--------------------------------------------------------------------+
| Caching Server 1 - IS NOT ON the DNS Whitelist |
| Caching Server 2 - IS ON the DNS Whitelist |
| Note: Transport between each host can be IPv4 or IPv6. |
+--------------------------------------------------------------------+
+----------+ +---------------+ +---------------+
| Stub | | DNS Caching | | DNS |
| Resolver | | Server 1 | | Server |
+----------+ +---------------+ +---------------+
| DNS Query: | |
| example.com A, AAAA | |
|---------------------->| |
| | |
| | DNS Query: |
| | example.com A, AAAA |
| |------------------------>|
| | |
| | | NOT on Whitelist
| | DNS Response: |
| | example.com A |
| |<------------------------|
| | |
| DNS Response: | |
| example.com A | |
|<----------------------| |
+----------+ +---------------+ +---------------+
| Stub | | DNS Caching | | DNS |
| Resolver | | Server 2 | | Server |
+----------+ +---------------+ +---------------+
| DNS Query: | |
| example.com A, AAAA | |
|---------------------->| |
| | |
| | DNS Query: |
| | example.com A, AAAA |
| |------------------------>|
| | |
| | | IS on Whitelist
| | DNS Response: |
| | example.com A, AAAA |
| |<------------------------|
| | |
| DNS Response: | |
| example.com A, AAAA | |
|<----------------------| |
DNS Resolver Whitelisting is functionally similar to Content Delivery Networks (CDNs) and Global Server Load Balancing (GSLB). When using a CDN or GSLB, a geographically-aware authoritative DNS server function is usually part of that overall system. As a result, the use of a CDN or GSLB with an authoritative DNS server function enables the IP address resource records returned to a resolver in response to a query to vary based on the estimated geographic location of the resolver [Wild-Resolvers] or a range of other technical factors. This CDN or GSLB DNS function is performed in order to attempt to direct hosts to connect to the nearest hosts (as measured in round trip time), to the host that has the best connectivity to an end user, to route around failures, to avoid sites where maintenance work has taken down hosts, and/or to the host that will otherwise provide the best service experience for an end user at a given point in time. As a result, one can see a direct similarity to DNS Resolver Whitelisting insofar as different IP address resource records are selectively returned to resolvers based on the IP address of each resolver and/or other imputed factors related to that IP address.
DNS Resolver Whitelisting has some similarities to DNS load balancing. There are of course many ways that DNS load balancing can be performed. In one example, multiple IP address resource records (A and/or AAAA) can be added to the DNS for a given FQDN. This approach is referred to as DNS round robin [RFC1794]. DNS round robin may also be employed where SRV resource records are used [RFC2782]. In another example, one or more of the IP address resource records in the DNS will direct traffic to a load balancer. That load balancer, in turn, may be application-aware, and pass the traffic on to one or more hosts connected to the load balancer which have different IP addresses. In cases where private IPv4 addresses are used [RFC1918], as well as when public IP addresses are used, those end hosts may not necessarily be directly reachable without passing through the load balancer first. So, similar to DNS Resolver Whitelisting, a load balancer will control what server host an end user's host communicates with when using a FQDN.
DNS Resolver Whitelisting has some similarities to so-called split DNS, briefly described in Section 3.8 of [RFC2775]. When split DNS is used, the authoritative DNS server selectively returns different responses depending upon what host has sent the query. While [RFC2775] notes the typical use of split DNS is to provide one answer to hosts on an Intranet (internal network) and a different answer to hosts on the Internet (external or public network), the basic idea is that different answers are provided to hosts on different networks. This is similar to the way that DNS Resolver Whitelisting works, whereby hosts on different networks which use different DNS recursive resolvers, receive different answers if one DNS recursive resolver is on the whitelist and the other is not. However, Internet transparency and Internet fragmentation concerns regarding split DNS are detailed in Section 2.1 of [RFC2956] and Section 2.7 notes concerns regarding split DNS and that it "makes the use of Fully Qualified Domain Names (FQDNs) as endpoint identifiers more complex". Section 3.5 of [RFC2956] further recommends that maintaining a stable approach to DNS operations is key during transitions, such as the one to IPv6 that is underway now, stating that "Operational stability of DNS is paramount, especially during a transition of the network layer, and both IPv6 and some network address translation techniques place a heavier burden on DNS."
While techniques such as GLSB and DNS load balancing, which share much in common with DNS Resolver Whitelisting and are widespread, some in the community have raised a range of concerns about the practice. Some concerns are specific DNS Resolver Whitelisting [WL-Concerns]. Other concerns are not as specific and pertain to the general practice of implementing content location or other network policy controls in the "middle" of the network in a so-called "middlebox" function. Whether such DNS-related functions are really part of a middlebox is debatable. Nevertheless, implementers should at least be aware of some of the risks of middleboxes, as noted in [RFC3724]. A related document, [RFC1958] explains that the state, policies, and other functions needed in the middle of the network should be minimized as a design goal. In addition, Section 2.16 of [RFC3234] makes specific statements concerning modified DNS servers. [RFC3234] also outlines more general concerns in Section 1.2 about the introduction of new failure modes when configuration is no longer limited to two ends of a session, so that diagnosis of failures and misconfigurations could become more complex. Two additional sources worth considering are [Tussle] and [Rethinking], in which the authors note concerns regarding the introduction of new control points (such as in middleboxes), including in the DNS.
However, some state, policies, and other functions have always been necessary to enable effective, reliable, and high-quality end-to-end communications on the Internet. In addition, techniques such as Global Server Load Balancing, Content Delivery Networking, DNS Load Balancing and Split DNS are not only widely deployed but are almost uniformly viewed as essential to the functioning of the Internet and highly beneficial to the quality of the end user experience on the Internet. These techniques have had and continue to have a beneficial effect on the experience of a wide range of Internet applications and protocols. So while there are valid concerns about implementing policy controls in the "middle" of the network, or anywhere away from edge hosts, the definition of what constitutes the middle and edge of the network is debatable in this case. This is particularly so given that GSLBs and CDNs facilitate connections from end host and the optimal content hosts, and could therefore be considered a modest and in many cases essential network policy extension of a network's edge, especially in the case of high-service-level domains.
There may be additional implications for end users that have configured their hosts to use a third party as their DNS recursive resolver, rather than the one(s) provided by their network operator. In such cases, it will be more challenging for a domain using whitelisting to determine the level of IPv6-related impairment when such third-party DNS recursive resolvers are used, given the wide variety of end user access networks which may be used and that this mix may change in unpredictable ways over time.
With DNS Resolver Whitelisting, DNS recursive resolvers can receive AAAA resource records only if they are on the whitelist. DNS Blacklisting is by contrast the the opposite of that, whereby all DNS recursive resolvers can receive AAAA resource records unless they are on the blacklist. Some implementers of DNS Resolver Whitelisting may choose to subsequently transition to DNS Blacklisting. It is unclear when and if it may be appropriate for a domain to change from whitelisting to blacklisting. Nor is it clear how implementers will judge the network conditions to have changed sufficiently to justify disabling such controls.
When a domain uses blacklisting, they are enabling all DNS recursive resolvers to receive AAAA record responses except for what is presumed to be a relatively small number that are on the blacklist. Over time it is likely that the blacklist will become smaller as the networks associated with the blacklisted DNS recursive resolvers are able to meaningfully reduce IPv6-related impairments to some acceptable level, though it is possible that some networks may never achieve that. DNS Blacklisting is also likely less labor intensive for a domain than performing DNS Resolver Whitelisting on a manual basis. This is simply because the domain would presumably be focused on a smaller number of DNS recursive resolvers with well known IPv6-related problems.
It is also worth noting that the email industry has a long experience with blacklists and, very generally speaking, blacklists tend to be effective and well received when it is easy to discover if an IP address is on a blacklist, if there is a transparent and easily understood process for requesting removal from a blacklist, and if the decision-making criteria for placing a server on a blacklist is transparently disclosed and perceived as fair. However, in contrast to an email blacklist where a blacklisted host cannot send email to a domain at all, with DNS Resolver Whitelisting communications will still occur over IPv4 transport.
As an alternative to adopting any of the aforementioned migration tactics, domains can choose to transition to native dual stack directly by adding native IPv6 capabilities to their network and hosts and by publishing AAAA resource records in the DNS for named resources within their domain. Of course, a domain can still control this transition gradually, on a name-by-name basis, by adding native IPv6 to one name at a time, such as mail.example.com first and www.example.com later. So even a "direct" transition can be performed gradually.
It is then up to end users with IPv6-related impairments to discover and fix any applicable impairments. However, the concerns and risks related to traffic volume Section 2.3 should still be considered and managed, since those are not directly related to such impairments. Not all content providers (or other domains) may face the challenges detailed herein or face them to the same degree, since the user base of each domain, traffic sources, traffic volumes, and other factors obviously varies between domains.
For example, while some content providers have implemented DNS Resolver Whitelisting (one migration tactic), others have run IPv6 experiments whereby they added AAAA resource records and observed and measured errors, and then decided not to implement DNS Resolver Whitelisting [Heise]. A more widespread such experiment was World IPv6 Day [W6D], sponsored by the Internet Society, on June 8, 2011. This was a unique opportunity for hundreds of domains to add AAAA resource records to the DNS without using DNS Resolver Whitelisting, all at the same time. Some of the participating domains chose to leave AAAA resource records in place following the experiment based on their experiences.
Content providers can run their own independent experiments in the future, adding AAAA resource records for a brief period of time (minutes, hours, or days), and then analyzing any impacts or effects on traffic and the experience of end users. They can also simply turn on IPv6 for their domain, which may be easier when the transition does not involve a high-service-level domain.
These usefulness of each tactic in Section 4, and the associated pros and cons associated with each tactic, is relative to each potential implementer and will therefore vary from one implementer to another. As a result, it is not possible to say that the potential phases below make sense for every implementer. This also means that the duration of each phase will vary between implementers, and even that different implementers may skip some of these phases entirely. Finally, the tactics listed in Section 4 are by no means exclusive.
In this phase, a site is accessible only via IPv4 transport. As of the time of this document, the majority of content on the Internet is in this state and is not accessible natively over IPv6.
One possible first step for a domain is to gain experience using a specialized new FQDN, such as ipv6.example.com or www.ipv6.example.com, as explained in Section 4.2.
As noted in Section 4.3, a domain could begin using DNS Resolver Whitelisting as a way to incrementally enable IPv6 access to content. This tactic may be especially interesting to high-service-level domains.
For a domain that decides to undertake DNS Resolver Whitelisting on a manual basis, the domain may subsequently move to perform DNS Resolver Whitelisting on an automated basis. This is explained in Section 4.3, and can significantly ease any operational burdens relating to a manually-maintained whitelist.
Domains that choose to implement DNS Resolver Whitelisting generally consider it to be a temporary measure. Many implementers have announced that they plan to permanently turn off DNS Resolver Whitelisting beginning on the date of the World IPv6 Launch, on June 6, 2012 [World IPv6 Launch]. For any implementers that do not turn off DNS Resolver Whitelisting at that time, it may be unclear how each and every one will judge when the network conditions to have changed sufficiently to justify turning off DNS Resolver Whitelisting. That being said, it is clear that the extent of IPv6 deployment to end users in networks, the state of IPv6-related impairment, and the maturity of IPv6 operations are all important factors. Any such implementers may wish to take into consideration that, as a practical matter, it will be impossible to get to a point where there are no longer any IPv6-related impairments; some reasonably small number of hosts will inevitably be left behind as end users elect not to upgrade them or as some hosts are incapable of being upgraded.
Regardless of whether a domain has first implemented DNS Resolver Whitelisting or has never done so, DNS Blacklisting as described in Section 4.4 may become interesting. This may be at the point in time when domains wish to make their content widely available over IPv6 but still wish to protect end users of a few networks with well known IPv6 limitations from having a bad end user experience.
A domain can arrive at this phase either following the use of a previous IPv6 migration tactic, or they may go directly to this point as noted in Section 4.5. In this phase the site's content has been made natively accessible via both IPv4 and IPv6 for all end users on the Internet, or at least without the use of any other IPv6 migration tactic.
If DNS Resolver Whitelisting is adopted, as noted in Section 4.3, then organizations which apply DNS Resolver Whitelisting policies in their authoritative servers should have procedures and systems which do not allow unauthorized parties to modify the whitelist or blacklist, just as all configuration settings for name servers should be protected by appropriate procedures and systems. Such unauthorized additions or removals from the whitelist can be damaging, causing content providers and/or network operators to incur support costs resulting from end user and/or customer contacts, as well as causing potential dramatic and disruptive swings in traffic from IPv6 to IPv4 or vice versa.
DNS security extensions defined in [RFC4033], [RFC4034], and [RFC4035] use cryptographic digital signatures to provide origin authentication and integrity assurance for DNS data. This is done by creating signatures for DNS data on a Security-Aware Authoritative Name Server that can be used by Security-Aware Resolvers to verify the answers. Since DNS Resolver Whitelisting is implemented on an authoritative DNS server, which provides different answers depending upon which DNS resolver has sent a query, the DNSSEC chain of trust is not altered. So even though an authoritative DNS server will selectively return AAAA resource records and/or A resource records, these resource records can certainly still be signed. In practical terms this means that two separate views or zones are used, each of which is signed, so that whether or not particular resource records exist, the existence or non-existence of the record can still be validated using DNSSEC. As a result, there should not be any negative impact on DNSSEC for those domains that have implemented both DNSSEC on their Security-Aware Authoritative Name Servers and also implemented DNS Resolver Whitelisting. As for any party implementing DNSSEC of course, such domains should ensure that resource records are being appropriately and reliably signed.
However, network operators that run DNS recursive resolvers should be careful not to modify the responses received from authoritative DNS servers. It is possible that some networks may attempt to do so in order to prevent AAAA record responses from going to end user hosts, due to some IPv6-related impairment or other lack of IPv6 readiness with that network. But when a network operates a Security-Aware Resolver, modifying or suppressing AAAA resource records for a DNSSEC-signed domain could break the chain of trust established with DNSSEC.
As noted in Section 4.1, there is a benefit in sharing IPv6-related impairment statistics within the Internet community over time. Any statistics that are shared or disclosed publicly should be aggregate statistics, such as "the domain example.com has observed an average daily impairment rate of 0.05% in September 2011, down from 0.15% in January 2011". They should not include information that can directly or indirectly identify individuals, such as names or email addresses. Sharing only aggregate data can help protect end user privacy and any information which may be proprietary to a domain.
In addition, there are often methods to detect IPv6-related impairments for a specific end user, such as running an IPv6 test when an end user visits the website of a particular domain. Should a domain then choose to automatically communicate the facts of an impairment to an affected user, there are likely no direct privacy considerations. However, if the domain then decided to share information concerning that particular end user with that user's network operator or another third party, then the domain may wish to consider advising the end user of this and seeking to obtain the end user's consent to share such information.
Appropriate guidelines for any information sharing likely varies by country and/or legal jurisdiction. Domains should consider any potential privacy issues when considering what information can be shared. If a domain does publish or share detailed impairment statistics, they would be well advised to avoid identifying individual hosts or users.
Finally, if a domain chooses to contact end userd directly concerning their IPv6 impairments, that domain should ensure that such communication is permissable under any applicable privacy policies of the domain or its websites.
There are situations where the differing quality of the IPv4 and IPv6 connectivity of an end user could cause complications in accessing content when a domain is using an IPv6 migration tactic. While today most end users' IPv4 connectivity is typically superior to IPv6 connectivity (if such connectivity exists at all), there could be implications when the reverse is true and and end user has markedly superior IPv6 connectivity as compared to IPv4. This is not a theoretical situation; it has been observed by at least one major content provider.
For example, in one possible scenario, a user is issued IPv6 addresses by their ISP and has a home network and devices or operating systems which fully support native IPv6. As a result this theoretical user has very good IPv6 connectivity. However, this end user's ISP has exhausted their available pool of unique IPv4 address, and uses NAT in order to share IPv4 addresses among end users. So for IPv4 content, the end user must send their IPv4 traffic through some additional network element (e.g. large scale NAT, proxy server, tunnel server). Use of this additional network element might cause an end user to experience sub-optimal IPv4 connectivity when certain protocols or applications are used. This user then has good IPv6 connectivity but impaired IPv4 connectivity. As a result, the user's poor IPv4 connectivity situation could potentially be exacerbated when accessing a domain which is using a migration tactic that causes this user to only be able to access content over IPv4 transport for whatever reason.
Should this sort of situation become widespread in the future, a domain may wish to take it into account when deciding how and when to transition content to IPv6.
There are no IANA considerations in this document.
The following people made significant textual contributions to this document and/or played an important role in the development and evolution of this document:
- John Brzozowski
- Chris Griffiths
- Tom Klieber
- Yiu Lee
- Rich Woundy
The author and contributors also wish to acknowledge the assistance of the following individuals or groups. Some of these people provided helpful and important guidance in the development of this document and/or in the development of the concepts covered in this document. Other people assisted by performing a detailed review of this document, and then providing feedback and constructive criticism for revisions to this document, or engaged in a healthy debate over the subject of the document. All of this was helpful and therefore the following individuals merit acknowledgement:
- Bernard Aboba
- Mark Andrews
- Jari Arkko
- Fred Baker
- Ron Bonica
- Frank Bulk
- Brian Carpenter
- Lorenzo Colitti
- Alissa Cooper
- Dave Crocker
- Ralph Droms
- Wesley Eddy
- J.D. Falk
- Adrian Farrel
- Stephen Farrell
- Tony Finch
- Karsten Fleischhauer
- Igor Gashinsky
- Wesley George
- Philip Homburg
- Jerry Huang
- Ray Hunter
- Joel Jaeggli
- Erik Kline
- Suresh Krishnan
- Victor Kuarsingh
- Marc Lampo
- Donn Lee
- John Leslie
- John Mann
- Danny McPherson
- Milo Medin
- Martin Millnert
- Russ Mundy
- Thomas Narten
- Pekka Savola
- Robert Sparks
- Barbara Stark
- Joe Touch
- Hannes Tschofenig
- Tina Tsou
- Members of the Broadband Internet Technical Advisory Group (BITAG)
| [World IPv6 Launch] | The Internet Society, "World IPv6 Launch Website", 2012. |
| [IPv6-Brokenness] | Anderson, T., "Measuring and Combating IPv6 Brokenness", Reseaux IP Europeens (RIPE) 61st Meeting, November 2010. |
| [Wild-Resolvers] | Ager, B., Smaragdakis, G., Muhlbauer, W. and S. Uhlig, "Comparing DNS Resolvers in the Wild", ACM Sigcomm Internet Measurement Conference 2010, November 2010. |
| [Heise] | Heise.de, "The Big IPv6 Experiment", Heise.de Website http://www.h-online.com, January 2011. |
| [W6D] | The Internet Society, "World IPv6 Day - June 8, 2011", Internet Society Website http://www.isoc.org, January 2011. |
| [WL-Ops] | Kline, E., "IPv6 Whitelist Operations", Google Google IPv6 Implementors Conference, June 2010. |
| [IPv6-Growth] | Colitti, L., Gunderson, S.H., Kline, E. and T. Refice, "Evaluating IPv6 adoption in the Internet", Passive and Active Management (PAM) Conference 2010, April 2010. |
| [WL-Concerns] | Brzozowski, J., Griffiths, C., Klieber, T., Lee, Y., Livingood, J. and R. Woundy, "IPv6 DNS Resolver Whitelisting - Could It Hinder IPv6 Adoption?", ISOC Internet Society IPv6 Deployment Workshop, April 2010. |
| [IETF-77-DNSOP] | Gashinsky, I., "IPv6 & recursive resolvers: How do we make the transition less painful?", IETF 77 DNS Operations Working Group, March 2010. |
| [NW-Article-DNSOP] | Marsan, C., "Yahoo proposes 'really ugly hack' to DNS", Network World , March 2010. |
| [NW-Article-DNS-WL] | Marsan, C., "Google, Microsoft, Netflix in talks to create shared list of IPv6 users", Network World , March 2010. |
| [Tussle] | Braden, R., Clark, D., Sollins, K. and J. Wroclawski, "Tussle in Cyberspace: Defining Tomorrow's Internet", Proceedings of ACM Sigcomm 2002, August 2002. |
| [Rethinking] | Blumenthal, M. and D. Clark, "Rethinking the design of the Internet: The end to end arguments vs. the brave new world", ACM Transactions on Internet Technology Volume 1, Number 1, Pages 70-109, August 2001. |
| [I-D.ietf-v6ops-happy-eyeballs] | Wing, D and A Yourtchenko, "Happy Eyeballs: Success with Dual-Stack Hosts", Internet-Draft draft-ietf-v6ops-happy-eyeballs-04, September 2011. |
| [RFC6343] | Carpenter, B., "Advisory Guidelines for 6to4 Deployment", RFC 6343, August 2011. |
[RFC Editor: This section is to be removed before publication]
-09: Minor updates to resolve questions in IETF Last Call
-08: Minor updates from v6ops WGLC
-07: Significant re-write based on feedback from Jari Arkko, Joel Jaeggli, Fred Baker, Igor Gashinsky, Donn Lee, Lorenzo Colitti, and Erik Kline.
-06: Removed the Open Issue #8 concerning the document name, at the direction of Joel Jaeggli. Removed Open Issue #2 from J.D. Falk and removed Open Issue #3 from Ray Hunter, as confirmed on the v6ops WG mailing list. Revised the Abstract and Intro as recommended by Tony Finch. Per Dave Crocker, updated the diagram following remedial ASCII art assistance, added a reference regarding IPv4-brokenness statistics. Removed Open Issue #1, after validating proper reference placement and removing NAT444 reference. Updates per Ralph Droms' review for the IESG. Closed Open Issue #4, Per Joe Touch, moved section 8 to just after section 3 - and also moved up section 6 and merged it. Closed Open Issue #5, per Dave Crocker and John Leslie, simplifying the document more, consolidating sections, etc. Closed Open Issue #6. Closed Open Issue #7, per Jari Arkko, ensuring all motivations are accounted for, etc. Closed Open Issue #9, per Stephen Farrell, re. World IPv6 Day (retained reference but re-worded those sections). Removed the happy-eyeballs reference since this was an informative reference and the draft could be delayed due to that dependency. ALL OPEN ITEMS ARE NOW CLOSED.
-05: Additional changes requested by Stephen Farrell intended to close his Discuss on the I-D. These changes were in Sections 6.2 and 8.3. Also shortened non-RFC references at Stephen's request.
-04: Made changes based on feedback received during IESG review. This does NOT include updated from the more general IETF last call - that will be in a -05 version of the document. Per Ralph Droms, change the title of 6.2 from "Likely Deployment Scenarios" to "General Implementation Variations", as well as changes to improve the understanding of sentences in Sections 2, 3, 4, and 8.2. Per Adrian Farrel, made a minor change to Section 3. Per Robert Sparks, to make clear in Section 2 that whitelisting is done on authoritative servers and not DNS recursive resolvers, and to improve Section 8.3 and add a reference to I-D.ietf-v6ops-happy-eyeballs. Per Wesley Eddy, updated Section 7.3.2 to address operational concerns and re-titled Section 8 from "Solutions" to "General Implementation Variations". Per Stephen Farrell, added text to Section 8.1 and Section 6.2, with a reference to 8.1 in the Introduction, to say that universal deployment is considered harmful. Added text to Section 2 per the v6ops list discussion to indicate that whitelisting is independent of the IP address family of the end user host or resolver. There was also discussion with the IESG to change the name of the draft to IPv6 DNS Resolver Whitelisting or IPv6 AAAA DNS Resolver Whitelisting (as suggested originally by John Mann) but there was not a strong consensus to do so. Added a new section 7.7, at the suggestion of Philip Homburg. Per Joe Touch, added a new Section 8.4 on blacklisting as an alternative, mentioned blacklisting in Section 2, added a new Section 7.8 on the use of 3rd party resolvers, and updated section 6.2 to change Internet Draft documents to RFCs. Minor changes from Barbara Stark. Changes to the Privacy Considerations section based on feedback from Alissa Cooper. Changed "highly-trafficked" domains to "high-traffic" domains. Per Bernard Aboba, added text noting that a whitelist may be manually or automatically updated, contrasting whitelisting with blacklisting, reorganized Section 3, added a note on multiple clearinghouses being possible. Per Pekka Savola, added a note regarding multiple clearinghouses to the Ad Hoc section, corrected grammar in Section 7.5, reworded Section 7.3.7, corrected the year in a RIPE reference citation. Also incorporated general feedback from the Broadband Internet Technical Advisory Group. Per Jari Arkko, simplified the introduction to the Implications section, played down possible impacts on RFC 4213, added caveats to Section 8.3.2 on the utility of transition names, re-wrote Section 9. Updated the Abstract and Introduction, per errors noted by Tony Finch. Updated the Security Considerations based on feedback from Russ Mundy. Per Ray Hunter, added some text to the De-Whitelisting implications section regarding effects on networks of switching from IPv6 to IPv4. Updated 7.3.1 per additional feedback from Karsten Fleischhauer. Per Dave Crocker, added a complete description of the practice to the Abstract, added a note to the Introduction that the operational impacts are particularly acute at scale, added text to Intro to make clear this practice affects all protocols and not just HTTP, added a new query/response diagram, added text to the Abstract and Introduction noting that this is an IPv6 transition mechanism, and too many other changes to list.
-03: Several changes suggested by Joel Jaeggli at the end of WGLC. This involved swapping the order of Section 6.1 and 6.2, among other changes to make the document more readable, understandable, and tonally balanced. As suggested by Karsten Fleischhauer, added a reference to RFC 4213 in Section 7.1, as well as other suggestions to that section. As suggested by Tina Tsou, made some changes to the DNSSEC section regarding signing. As suggested by Suresh Krishnan, made several changes to improve various sections of the document, such as adding an alternative concerning the use of ipv6.domain, improving the system logic section, and shortening the reference titles. As suggested by Thomas Narten, added some text regarding the imperfection of making judgements as to end user host impairments based upon the DNS recursive resolver's IP and/or network. Finally, made sure that variations in the use of 'records' and 'resource records' was updated to 'resource records' for uniformity and to avoid confusion.
-02: Called for and closed out feedback on dnsop and v6ops mailing lists. Closed out open feedback items from IETF 79. Cleared I-D nits issues, added a section on whether or not this is recommended, made language less company-specific based on feedback from Martin Millnert, Wes George, and Victor Kuarsingh. Also mentioned World IPv6 Day per Wes George's suggestion. Added references to the ISOC World IPv6 Day and the Heise.de test at the suggestion of Jerry Huang, as well as an additional implication in 7.3.7. Made any speculation on IPv4 impairment noted explicitly as such, per feedback from Martin Millnert. Added a reference to DNS SRV in the load balancing section. Added various other references. Numerous changes suggested by John Brzozowski in several sections, to clean up the document. Moved up the section on why whitelisting is performed to make the document flow more logically. Added a note in the ad hoc deployment scenario explaining that a deployment may be temporary, and including more of the perceived benefits of this tactic. Added a Privacy Considerations section to address end-user detection and communication.
-01: Incorporated feedback received from Brian Carpenter (from 10/19/2010), Frank Bulk (from 11/8/2010), and Erik Kline (from 10/1/2010). Also added an informative reference at the suggestion of Wes George (from from 10/22/2010). Closed out numerous editorial notes, and made a variety of other changes.
-00: First version published as a v6ops WG draft. The preceding individual draft was draft-livingood-dns-whitelisting-implications-01. IMPORTANT TO NOTE that no changes have been made yet based on WG and list feedback. These are in queue for a -01 update.
[RFC Editor: This section is to be removed before publication]
Check references to ensure all of them are still necessary