| Network Working Group | S. Barber |
| Internet-Draft | Cox Communications |
| Intended status: Informational | O. Delong |
| Expires: January 05, 2012 | Hurricane Electric |
| C. Grundemann | |
| CableLabs | |
| V. Kuarsingh | |
| Rogers Communications | |
| B. Schliesser | |
| Cisco Systems | |
| July 04, 2011 |
ARIN Draft Policy 2011-5: Shared Transition Space
draft-bdgks-arin-shared-transition-space-00
This memo encourages the reservation of a Shared Transition Space, an IPv4 prefix designated for local use within service provider networks during the period of IPv6 transition. This address space has been proposed at various times in the IETF, and more recently by the ARIN policy development community.
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 January 05, 2012.
Copyright (c) 2011 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.
As the Internet community approaches exhaustion of unallocated IPv4 numbers, the value of globally unique addresses is becoming manifest. More than ever network operators recognize the need to transition to the IPv6 address family. However, the immediate necessity of continued IPv4 connectivity poses a near-term challenge - without adequate IPv4 resources, most network operators must deploy more efficient addressing architectures and many must deploy address-sharing technologies.
In order to facilitate these operators' need for near-term IPv4 connectivity, [I-D.weil-shared-transition-space-request] proposes the reservation of a /10 IPv4 prefix for use in Service Provider (SP) networks. Referred to as Shared Transition Space, this address block would facilitate SP deployment of non-unique address plans that do not conflict with traditional Private [RFC1918] address space. By using the Shared Transition Space operators may deploy CGN [I-D.ietf-behave-lsn-requirements] internal networks, extranet [RFC4364] communities, and/or SP-local services without consuming globally unique addresses.
However, given the Feb 2011 depletion of the IANA Free Pool inventory [NRO-IANA-exhaust] it is not currently possible for the IANA to reserve an IPv4 /10 prefix as recommended in [I-D.weil-shared-transition-space-request]. Thus the ARIN community has proposed in Draft Policy [ARIN-2011-5] the reservation of a Shared Transition Space from the ARIN inventory of unallocated IPv4 numbers. After much discussion by the ARIN community, [ARIN-2011-5] was recommended by the ARIN Advisory Council for approval by the ARIN Board of Trustees.
Essentially similar to [I-D.weil-shared-transition-space-request], Draft Policy [ARIN-2011-5] is currently pending ARIN Board approval. The ARIN Board has asked for IAB clarification with regard to responsibilities outlined in [RFC2860] and has received a response [IAB-response] indicating that the IETF holds responsibility for the reservation of specialized address blocks. Thus, this memo is a discussion of the merits of a Shared Transition Space, and is a call for consensus between the IETF and RIR communities that such an address reservation should be made.
There are a number of use-cases for the Shared Transition Space. This section discusses the primary scenarios envisioned at the time of this writing.
A primary use-case for the Shared Transition Space will be deployment in CGN internal networks, as described in [I-D.ietf-behave-lsn-requirements]. A key benefit of CGN is the ability to share a smaller number of Globally Unique Addresses (GUA) amongst a larger number of end-sites.
In one CGN deployment scenario sometimes referred to as NAT444 [I-D.shirasaki-nat444-isp-shared-addr], the CGN internal network is numbered with IPv4 addresses that are not globally routed while the end-sites are numbered with Private [RFC1918] addresses. In this scenario the Shared Transition Space will be used to provide contextually unique IPv4 addresses to end-site CPE devices and intermediate infrastructure.
Another use-case for the Shared Transition Space is in building private Extranet community networks. In these networks, multiple end-sites administered by different organizations are connected together via VPN technology. Because different organizations may be using Private address space internally, an Extranet addressing plan is often unable to effectively use Private address space without conflicting. The Shared Transition Space will provide an alternative to the use of GUA space in such a scenario.
In networks that contain local services (such as nameservers, content repositories or caches, etc) the Shared Transition Space will offer an alternative to GUA. For instance, video content servers that are available only to customers directly connected to the SP network might be addressed from the Shared Transition Space, preserving GUA for services that require global connectivity.
Many service providers have deployed a hierarchical network using Private [RFC1918] space, which has served them well for many years. Due in large part to the explosive growth of new services they have run out of available private space. While it is possible to re-engineer internal networks, such activity is customer impacting and operationally complex. Making more private space available for service providers allows for a manageable transition to IPv6 without significant impact to customers.
A number of possible alternatives to Shared Transition Space have been proposed and/or discussed by the Internet community. See, for instance, [I-D.azinger-additional-private-ipv4-space-issues] for a discussion of alternatives and potential issues. This section outlines these possible alternatives and briefly discusses their applicability.
Every discussion of the Shared Transition Space begins with an assumption that Globally Unique Addresses (GUA) are a preferable choice for numbering. This is almost always technically true. However, given the fundamental driver of IPv4 address exhaustion, GUA is not a pragmatic alternative to the Shared Transition Space.
Additionally, if various organizations use various GUA ranges to number CGN zones, it will be difficult for other networks and/or systems to deterministically know if the endpoints are using true internet reachable IPs, or if the source network may be using them as CGN zone space. This situation would likely lead to additional technical issues during various leakage conditions, filter rule issues (routing) and for CDN or other third party providers who may be present within the source network, to name a few.
In each of the use-cases for Shared Transition Space, it may be possible to instead use Private [RFC1918] address space. In situations where all endpoints in the network are managed by a single organization, this may be a viable option. However when end-sites are administered by different organizations and/or individuals, the possibility of address conflict becomes a significant risk to operations.
A study of DNS traffic [v6ops-msg06187] has shown that effectively all of the existing Private [RFC1918] address space is currently being used by end-sites attached to the Internet. While individual network environments may vary in this regard, most SP operators face the risk that their use of Private address space will conflict with their customer end-sites.
In the event of conflict, it is possible that the end-site CPE will fail and/or not function correctly. Some CPE implementations are known to support overlapping addresses on the "inside" and "outside" interfaces, however many others are known to fail under such circumstances. For SP operators, the Shared Transition Space offers a less risky alternative to GUA that retains the benefit of non-conflict.
Also, the use of Private [RFC1918] address space on interfaces and hosts often causes default behaviours on such hosts which may not be desirable when the endpoint is actually connected to the Internet. There are often behavioural expectations for Internet connected endpoints, regardless of them being subject to a NAT.
Incorrect affiliation of the WAN side interface being in a "protected" zone and/or on a trusted network may not be desirable. With NAT444 deployments, it is important that the endpoint (i.e. CPE) behave like any other internet node. One example of this from our testing was observed behaviours where some CPEs did not filter and/or firewall correctly when Private [RFC1918] address space was used on both WAN and LAN interfaces.
One proposed alternative to Shared Transition Space is the re-classification and use of the 240.0.0.0/4 "Class E" address space as unicast. This has been proposed, for instance, by [I-D.fuller-240space] and [I-D.wilson-class-e]. While this alternative might be possible in tightly constrained environments, where all of the network elements are known to support Class E address space, it is not generally useful in the use-cases described above. At this time, a significant number of IPv4 stack implementations treat the Class E address space as reserved and will not route, forward, and/or originate traffic for that range.
An unfortunate alternative to the Shared Transition Space is "prefix squatting", in which the operator re-uses another organization's IPv4 allocation for their own numbering needs. When this approach results in the other organization's prefix being announced globally by the "squatting" operator, it is often referred to as "prefix hijacking". However, this discussion is focused on scenarios in which the prefix is not announced globally but is, rather, used for internal numbering only.
In this scenario, the allocation may not be routed globally by the legitimate address holder, making it attractive for such purposes. Or it may be routed but "uninteresting" to the SP network's endpoints. In either case there is a potential for conflict in the event that any end-site actually wishes to communicate with the legitimate address holder. As such, this alternative is to be discouraged with prejudice.
It is important to note that there are no behavioural advantages to using "squat space" over using assigned "shared space". Both options subject the CPE to the same general behaviours (GUA space, but not globally reachable). The only real difference is the negative impacts of squatting (as noted above) and the advantages of a community coordinated and standardized prefix.
The primary reason that any network would be likely to adopt "prefix squatting" is if they are faced with the operational realities of CGN before/without the allocation of a shared transition space.
Similar to "Prefix Squatting" but significantly less dangerous, this alternative involves the reuse by an operator of their own address allocations. In this scenario, a network operator might use the same prefix for multiple "regions" and/or extranet communities. For instance, in CGN deployments the operator might reuse the same GUA prefix across multiple geographic regions (e.g. without announcing it globally).
Here again, it is important to note that there are no behavioural advantages gained over a "shared space" but there is the added community cost of each network having to dedicate a unique block of addresses to this purpose, consuming far more resources than a single block of "shared space".
In the event that the Internet community doesn't set aside an IPv4 prefix for Shared Transition Space, it is possible that a number of SP operators can come together and designate an address block to be "shared" amongst them for an identical purpose. This would have the same technical merits as an IETF and/or RIR sponsored Shared Transition Space, however it would lack the efficiency of a community coordinated and standardized prefix for such purposes, gain no behavioural advantages, remove the deterministic nature of managing a single range and also subjects the Internet (users of the space) to additional risk since any member of the consortium who has contributed space could later pull out and potentially cause disruptions in multiple networks.
Availability of a Shared Transition Space helps SPs continue to meet the demands of IPv4 address and/or connectivity post exhaustion. For environments where CGN in a NAT444 scenario is necessary, addresses from this space can be used to provide intermediary network addressing assisting in provided IPv4 flow continuity for new or migrating customers.
In other circumstances, the shared transition space allows SPs to number devices in the network which do not require globally reachablity without the need for fulfillment thorough an RIR.
If operators are required to us their individually allocated GUA where "shared space" would have applied, they will face exhaustion sooner and thus be forced to deploy CGN sooner as well. Operators can postpone this deployment of CGN by using "shared space" for internal uses, because that allows more efficient use of their remaining GUA in places where global uniqueness is truly mandatory.
The shared transition space can also be used to number and reclaim IPv4 addresses within provider networks which do not require global reachability. This option can be used by many networks worldwide, it provides an option for using currently assigned space much more efficiently.
Operators can re-deploy recovered addresses for customers that need them (including new / static / GUA customers), hosted servers, etc. or to facilitate other efforts that might provide even more efficient use of GUA space within the network. The freed addresses can be assigned to endpoints which require IPv4 global reachablity and thus help delay and/or remove the need for CGN.
In cases where the operator doesn't need the recovered addresses, they can be made available to others that do need them. This may be through voluntary return the RIR, or through transfer to another network operator (perhaps via a market mechanism).
If an SP determines that the space recovered is not needed, the space can be returned or transferred through those mechanisms already in place with the RIRs. For example, in the ARIN region, there are such mechanisms already defined in the ARIN NRPM section 8.3 [ARIN-NRPM-8.3].
While making "shared space" available to the community, may or may not lessen the demand on the RIRs for allocations, it will help ensure that the address resources which remain in inventory are used most efficiently, maximizing the use of that inventory for services that require global routability.
BENSON: note that I changed this to "may or may not" because I think it's arguable either way... If an SP doesn't need globally unique space to continue numbering (with a CGN) then that might slow the rate of allocation. On the other hand, unless RIRs police it, the SP's interest is aligned with continuing to make requests and "bank" them for later, so that wouldn't change the rate of allocation at all.
Standardizing on a single block will help the community develop standard ways of selecting, routing, filtering and managing shared space. This task would be much more difficult or impractical for any of the alternative options.
Standard internal routing policy and filtering can be applied uniformly inside network environments. Additionally, exchange points between networks can have standard policies applied allowing operators to protect each other from CGN zone IPs leaking between networks. This may not be possible with squat space since many operators will not divulge what space may be used and with Private [RFC1918] address space where each operator may only be able to free up certain portions of the space which are not likely to be consistent between networks.
Operators will need to grapple with the need to provide IPv4 based flow continuity to customers post exhaustion. By removing the burden of operators needing to find adequate IPv4 address space to meet the needs that a shared transition space can fulfill, they can concentrate on the real task at hand: Deploying IPv6.
Endless cycles can be avoided where operators squat, free up space and/or segment networks in an effort to find valid IPv4 space. The saved effort, time and cost can be re-directed to provided quality IPv6 connectivity therefore expiditing the removal of the overall need for IPv4 addresses and connectivity.
NAT is understood to be less then optimal [RFC6269], especially when implemented as CGN [I-D.donley-nat444-impacts]. That said, it is a necessary technology for many networks and cannot be completely avoided. Since the number of IPv4 internet endpoints will exceed the number of IPv4 addresses which are available for Internet connectivity, NATs are needed.
While the authors agree that "NAT is bad", it must also be understood that shared transition space does not change the fundamental problems with NAT and so those problems will not be discussed at length here.
The use of GUA space (non-RFC1918) does in some circumstances break some functions. The most commonly cited function is 6to4. Although 6to4 can break, it's not commonly turned on by default. 6to4 can also be "repaired" in some instances when used behind a CGN (NAT66) or managed by the network operator. Since the volume of impacted endpoints will be very low, operators can likely manage the disabling of 6to4 when needed.
The value of a Shared Transition Space may be diminished if misused by end-sites as generic Private addresses. Thus, the reservation must be clearly designated for use by SPs that are providing infrastructure as described herein.
This is not a technical issue. As with any technology, the opportunity exists for a misuse. This however should not shroud the strong benefits of the shared address space option. Many technologies in use today can be used correctly or misused. This does not prevent the community from introducing those technologies since the good far outweighs the bad.
As an example, the use of DiffServ [RFC2475] can result in punitive measures for some hosts in a network while favouring others bad on illegitimate rules. This however is not a good argument on why not to permit DiffServ.
This argument is simply incorrect. Post IPv4-exaustion, any SP that wishes to continue providing IPv4 connectivity will necessarily deploy network architectures and technologies that require such an address space.
In absense of a designated Shared Transition Space, operators will use GUA space in essentially the same ways described in this memo, with or without IETF acknowledgement. The industry needs to recognize this and work in the best interests of the "real customer".
While customer growth for some ISPs has slowed, for many service providers new services are growing at a faster rate than has been anticipated. Wireline voice customers for example require two-way communication paths to allow them to function properly. IP enabled televisions is another example of devices that support video and voice services and require IP addresses. In many cases the protocols that allow these devices to work have embedded IP addresses that do not work with NAT. The only way to maintain these services, which in many cases are considered lifeline, is to provide them with an IP address that is unique with the service provider network.
Likewise, growth continues to exist in some geographical regions. While some areas have slower growth, as a result of significant penetration of Internet access, there are still many areas with unmet needs, growing populations, or both.
With the IANA inventory essentially exhausted [NRO-IANA-exhaust] it is only a matter of time before each of the RIRs are unable to satisfy requests for IPv4 addresses. [GIH-When] In fact, the APNIC has already allocated all but their final /8 of inventory [APNIC-final-slash8] and is no longer making allocations larger than a /22 prefix. Each of the other RIRs is on a trajectory toward exhaustion in the near future.
While some SPs have existing inventory that will outlast the RIR inventories, this is not universally true. In fact, the distribution of IPv4 number resources amongst operators is highly variable (based on size, history, etc) and in the worst cases is already becoming problematic.
While true that the Shared Transition Space will consume some unallocated inventory, the impact is no greater than would be seen if individual SPs continue to request allocations of GUA for the scenarios described herein. It is possible, rather, that the Shared Transition Space might alleviate some near-term demand on RIR inventories. However, even if the RIR inventories are exhausted at the current rate, the reservation of a Shared Transition Space will enable continued deployment of IPv4 connectivity by SP networks - a clear benefit.
There are many ISP networks that may require greater or lesser amounts of IPv4 number resources, as a Shared Transition Space. While a larger prefix (such as a /8) would allow for expanded applicability, to larger ISP networks, it is generally thought that a /10 will be adequate for a large number of network deployments. Likewise, a /10 seems to be appropriate given the current technological constraints and operational considerations of CGN deployment. On the other hand, a smaller prefix might not be large enough to apply in many modern network deployments. Thus, a /10 prefix for Shared Transition Space is considered an appropriate compromise.
It remains to be seen whether the reservation of a Shared Transition Space will delay the impending exhaustion of RIRs' IPv4 inventory. Certainly, the availability of this Shared Transition Space will satisfy a number of demands that would otherwise become requests for GUA resources. However, whether this translates to an actual reduction in requests is up to the RIRs and requesting organizations.
Although IPv6 is the strategic long term answer fro IPv4 address exhaustion, it does not immediately solve IPv4 connectivity requirements. There is an entire eco-system which exists on the Internet which is not IPv6 ready at this time. IPv4 flow continuity will be required for a long period of time.
Many businesses have long procurement and fulfilment cycles which will need to be used to upgrade networks to support IPv6. Also, the consumer (home) space is years away from being all IPv6 capable. Many homes are filled with IPv4 only consumer electronics, computers, TVs, accessories and other systems.
There are still a number of products that are either not IPv6 compliant, or for which the necessary criteria for being "IPv6 compliant" is unclear or undefined. Some examples include security products (IDS/IPS in particular), a large number of software applications (MySQL is one example), and there are still production systems (both inside companies and as products) being rolled out that are not IPv6 aware (until very recently, this included all Linksys routers).
The whole Internet needs to get to IPv6 more or less at the same time in order to avoid significant deployment of transition technologies. This proposal may help delay some transition technology deployment while IPv6 deployments move ahead. More IPv6 should mean less transition technology.
Although IPv6 is the strategic long term answer for IPv4 address exhaustion, it does not immediately solve IPv4 connectivity requirements. There is an entire eco-system which exists on the Internet today and is not IPv6 ready at this time. IPv4 flow continuity will be required for a long period of time.
Many businesses have long procurement and fulfilment cycles which will need to be used to upgrade networks to support IPv6. Also, the consumer (home) space is years away from being fully IPv6 capable. Many homes are filled with IPv4-only consumer electronics, computers, TVs, accessories and other systems.
BENSON: Note: "cite arkkoís drafts about operating in IPv6-only mode for evidence that this is not actually workable yet"
The proposal for additional Private space in order to survive IPv6 transition dates back to [I-D.hain-1918bis] in April 2004, and more recently as Shared Transition Space [I-D.shirasaki-isp-shared-addr] in June 2008 where a proposal to set aside "ISP Shared Space" has been made. During discussion of the more recent proposals many of the salient points where commented on including challenges with RFC1918 in the ISP NAT Zone, Avoidance of IP Address Conflicts, and challenges with 240/4.
This effort was followed up by [I-D.weil-opsawg-provider-address-space] in July 2010 which was re-worked in November 2010 as [I-D.weil-shared-transition-space-request]. The latter two efforts continued to point out the operators need for Shared Transition Space, with full acknowledgement that challenges may arise with NAT444 as per [I-D.donley-nat444-impacts].
Most recently, following exhaution of the IANA's /8 pool [NRO-IANA-exhaust], this proposal has been discussed by various RIR policy participants. As described herein, the body of ARIN policy development participants has recommended a Shared Address Space policy for adoption [ARIN-2011-5].
This history has shown that operators and other industry contributors have identified the need for a Shared Transition Space assignment based on clearly identified reasons. The previous work has also described the awareness of the challenges of this space, but points to this option as the most manageable and workable solution for IPv6 transition space.
The following is a brief history of ARIN Draft Policy 2011-5.
Draft Policy ARIN-2011-5
Shared Transition Space for IPv4 Address Extension
Date: 20 January 2011
Policy statement:
Updates 4.10 of the NRPM:
A second contiguous /10 IPv4 block will be reserved to facilitate IPv4 address extension. This block will not be allocated or assigned to any single organization, but is to be shared by Service Providers for internal use for IPv4 address extension deployments until connected networks fully support IPv6. Examples of such needs include: IPv4 addresses between home gateways and NAT444 translators.
Rationale:
The Internet community is rapidly consuming the remaining supply of unallocated IPv4 addresses. During the transition period to IPv6, it is imperative that Service Providers maintain IPv4 service for devices and networks that are currently incapable of upgrading to IPv6. Consumers must be able to reach the largely IPv4 Internet after exhaustion. Without a means to share addresses, people or organizations who gain Internet access for the first time, or those who switch providers, or move to another area, will be unable to reach the IPv4 Internet.
Further, many CPE router devices used to provide residential or small-medium business services have been optimized for IPv4 operation, and typically require replacement in order to fully support the transition to IPv6 (either natively or via one of many transition technologies). In addition, various consumer devices including IP-enabled televisions, gaming consoles, medical and family monitoring devices, etc. are IPv4-only, and cannot be upgraded. While these will eventually be replaced with dual-stack or IPv6 capable devices, this transition will take many years. As these are typically consumer-owned devices, service providers do not have control over the speed of their replacement cycle. However, consumers have an expectation that they will continue to receive IPv4 service, and that such devices will continue to have IPv4 Internet connectivity after the IPv4 pool is exhausted, even if the customer contracts for new service with a new provider.
Until such customers replace their Home Gateways and all IPv4-only devices with IPv6-capable devices, Service Providers will be required to continue to offer IPv4 services through the use of an IPv4 address sharing technology such as NAT444. A recent study showed that there is no part of RFC1918 space which would not overlap with some IPv4 gateways, and therefore to prevent address conflicts, new address space is needed.
Service providers are currently presented with three options for obtaining sufficient IPv4 address space for NAT444/IPv4 extension deployments: (1) Request allocations under the NRPM; (2) share address space with other providers (this proposal); or (3) use address space allocated to another entity (i.e. 'squat'). Of the three options, option 2 (this proposal) is preferable, as it will minimize the number of addresses used for IPv4 extension deployments while preserving the authority of IANA and RIRs.
Timetable for implementation: immediately
The authors would like to thank Jeffrey Finkelstein for his significant contributions.
The authors would also like to thank Chris Donley, Wes George, and Richard Von Scherr for their review, comments, and support.
This memo includes no request to IANA.
This memo makes reference to a number of deployment scenarios that have unique security considerations, and the reader is advised to investigate these independently.
While this memo does not introduce any specific technical issues that may be subject to detailed security considerations, it does reccommend the reservation of a new IPv4 address space that might have unique properties when deployed. As such, all implementors of this Shared Transition Space are encouraged to consider carefully the best practices associated with the use of this space, including considerations relating to filtering, routing, etc.