| Internet Engineering Task Force | W. George |
| Internet-Draft | Time Warner Cable |
| Intended status: Standards Track | S. Amante |
| Expires: April 2, 2015 | Apple, Inc. |
| September 29, 2014 |
Autonomous System Migration Features and Their Effects on the BGP AS_PATH Attribute
draft-ietf-idr-as-migration-03
This draft discusses some BGP features for ASN migration that, while commonly used, are not formally part of the BGP4 protocol specification and may be vendor-specific in exact implementation. It is necessary to document these de facto standards to ensure that they are properly supported in future BGP protocol work such as BGPSec.
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 April 2, 2015.
Copyright (c) 2014 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 draft discusses some BGP features for ASN migration that, while commonly used, are not formally part of the BGP4 [RFC4271] protocol specification and may be vendor-specific in exact implementation. These features are local to a given BGP Speaker and do not require negotiation with or cooperation of BGP neighbors. The deployment of these features do not need to interwork with one another to accomplish the desired results, so slight variations between existing vendor implementations exist, and will not necessarily be harmonized due to this document. However, it is necessary to document these de facto standards to ensure that new implementations can be successful, and any future protocol enhancements to BGP that propose to read, copy, manipulate or compare the AS_PATH attribute can do so without inhibiting the use of these very widely used ASN migration features.
The migration features discussed here are useful to ISPs and organizations of all sizes, but it is important to understand the business need for these features and illustrate why they are so critical for ISPs' operations. During a merger, acquisition or divestiture involving two organizations it is necessary to seamlessly migrate both internal and external BGP speakers from one ASN to a second ASN. The overall goal in doing so is to simplify operations through consistent configurations across all BGP speakers in the combined network. In addition, it is common practice in the industry for ISPs to bill customers based on utilization. ISPs bill customers based on the 95th percentile of the greater of the traffic sent or received, over the course of a 1-month period, on the customer's access circuit. Given that the BGP Path Selection algorithm selects routes with the shortest AS_PATH attribute, it is critical that the ISP does not increase AS_PATH length during or after ASN migration toward downstream transit customers or settlement-free peers, who are likely sending or receiving traffic from those transit customers. This would not only result in sudden changes in traffic patterns in the network, but also substantially decrease utilization driven revenue at the ISP.
By default, the BGP protocol requires an operator to configure a router to use a single remote ASN for the BGP neighbor, and the ASN must match on both ends of the peering in order to successfully negotiate and establish a BGP session. Prior to the existence of these migration features, it would have required an ISP to coordinate an ASN change with, in some cases, tens of thousands of customers. In particular, as each router is migrated to the new ASN, to avoid an outage due to ASN mismatch, the ISP would have to force all customers on that router to change their router configurations to use the new ASN immediately after the ASN change. Thus, it becomes critical to allow the ISP to make this process a bit more asymmetric, so that it could seamlessly migrate the ASN within its network(s), but allow the customers to gradually migrate to the ISP's new ASN at their leisure, either by coordinating individual reconfigurations, or accepting sessions using either the old or new ASN to allow for truly asymmetric migration.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119].
This draft uses Autonomous System Numbers (ASNs) from the range reserved for documentation as described in RFC 5398 [RFC5398]. In the examples used here, they are intended to represent Globally Unique ASNs, not private use ASNs as documented in RFC 6996 [RFC6996] section 5.
The use case being discussed here is an ISP merging two or more ASNs, where eventually one ASN subsumes the other(s). In this use case, we will assume the most common case where there are two ISPs, A and B, that prior to the ASN migration use AS 64500 and 64510, respectively. AS 64500 will be the permanently retained ASN used across the consolidated set of both ISPs network equipment, and AS 64510 will be retired. Thus, at the conclusion of the ASN migration, there will be a single ISP A' with all internal BGP speakers configured to use AS 64500. To all external BGP speakers, the AS_PATH length will not be increased.
In this same scenario, AS 64496 and AS 64499 represent two separate customer networks: C and D, respectively. Originally, customer C (AS 64496) is attached to ISP B, which will undergo ASN migration from AS 64510 to AS 64500. Furthermore, customer D (AS 64499) is attached to ISP A, which does not undergo ASN migration since the ASN for ISP A will remain constant, (AS 64500). Although this example refers to AS 64496 and 64499 as customer networks, either or both may be settlement-free or other types of peers. In this use case they are referred to as "customers" merely for convenience.
------ ------
/ ISP A \ / ISP B \
| AS 64500 | | AS 64510 |
\ / \ /
------- -------
| |
| |
------------ -------------
| Cust D | | Cust C |
| AS 64499 | | AS 64496 |
------------ -------------
Figure 1: Before Migration
---------------
/ \
| ISP A' |
| AS 64500 |
\ /
---------------
/ \
/ \
| |
------------ -------------
| Cust D | | Cust C |
| AS 64499 | | AS 64496 |
------------ -------------
Figure 2: After Migration
The general order of operations, typically carried out in a single maintenance window by the network undergoing ASN migration (ISP B), are as follows. First, ISP B will change the global BGP ASN used by a Provider Edge (PE) router, from ASN 64510 to 64500. At this point, the router will no longer be able to establish eBGP sessions toward the existing Customer Edge (CE) devices that are attached to it and still using AS 64510. Second, since ISP B needs to do this without coordinating the simultaneous change of its ASN with all of its eBGP peers, ISP B will configure two separate, but related ASN migration features discussed in this document on all eBGP sessions toward all CE devices. These features enable the router to establish BGP neighbors using the legacy ASN, modify the AS_PATH attribute received from a CE device when advertising it further, and modify AS_PATH when transmitted toward CE devices to achieve the desired effect of not increasing the length of the AS_PATH.
At the conclusion of the ASN migration, the CE devices at the edge of the network are not aware of the fact that their upstream router is now in a new ASN and do not observe any change in the length of the AS_PATH attribute. However, after the changes discussed in this document are put in place by ISP A', there is a change to the contents of the AS_PATH attribute to ensure the AS_PATH is not artificially lengthened while these AS migration parameters are used.
In this use case, neither ISP is using BGP Confederations RFC 5065 [RFC5065] internally.
There are multiple implementations with equivalent features deployed and in use. Some documentation pointers to these implementations, as well as additional documentation on migration scenarios can be found in the appendix. The examples cited below use Cisco IOS CLI for ease of illustration purposes only.
The following section addresses features that are specific to modifying the AS_PATH attribute at the Autonomous System Border Routers (ASBRs) of an organization, (typically a single Service Provider). This ensures that external BGP customers/peers are not forced to make any configuration changes on their CE routers before or during the exact time the Service Provider wishes to migrate to a new, permanently retained ASN. Furthermore, these features eliminate the artificial lengthening of the AS_PATH both transmitted from and received by the Service Provider that is undergoing AS Migration, which would have negative implications on path selection by external networks.
The first feature used in the process described above is called "Local AS". This feature allows the PE router that was formerly in ISP B to re-establish an eBGP session toward the existing CE devices using the legacy AS, AS 64510. Ultimately, the CE devices (i.e.: customer C) are completely unaware that ISP B has reconfigured its router to participate as a member of a new AS. Within the context of the former ISP B PE router, the second effect this feature has on AS_PATH is that, by default, it prepends all received BGP UPDATEs with the legacy AS of ISP B: AS 64510, while advertising it (Adj-RIB-Out) to other BGP speakers (A'). Within Loc-RIB on ISP B prior to the migration, the AS_PATH toward customer C would appear as: 64510, whereas the same RIB on ISP A' (ISP B routers post-migration) would contain AS_PATH: 64510 64496. To avoid changes to the AS_PATH length, a secondary feature "No Prepend" is added to the "Local AS" configuration toward every eBGP neighbor on PE routers migrating from ISP B. The "No Prepend" feature causes those routers to not prepend the legacy AS, AS 64510, when advertising UPDATES received from customer C. This restores the AS_PATH within ISP A' toward customer C so that it is just one ASN in length: 64496.
In the direction of CE -> PE (inbound):
PE-B is a PE that was originally in ISP B, and has a customer peer CE-B. PE-B has had its global configuration ASN changed from AS 64510 to AS 64500 to make it part of the permanently retained ASN. This now makes PE-B a member of ISP A'. PE-A is a PE that was originally in ISP A, and has a customer peer CE-A. Although its global configuration ASN remains AS 64500, throughout this exercise we also consider PE-A a member of ISP A'.
ISP A' ISP A'
CE-A <--- PE-A <------------------- PE-B <--- CE-B
64499 New_ASN: 64500 Old_ASN: 64510 64496
New_ASN: 64500
Note: Direction of BGP UPDATE as per the arrows.
Figure 3: Local AS BGP UPDATE Diagram
The final configuration on PE-B after completing the "Local AS" portion of the AS migration is as follows:
router bgp 64500
neighbor <CE-B_IP> remote-as 64496
neighbor <CE-B_IP> local-as 64510 no-prepend
As a result of the "Local AS No Prepend" configuration, on PE-B, CE-A will see an AS_PATH of: 64500 64496. CE-A will not receive a BGP UPDATE containing AS 64510 in the AS_PATH. (If only the "local-as 64510" feature was configured without the keyword "no-prepend" on PE-B, then CE-A would see an AS_PATH of: 64496 64510 64500, which results in an unacceptable lengthening of the AS_PATH).
The previous feature, "Local AS No Prepend", was designed to modify the AS_PATH Attribute received by the ISP in updates from CE devices, when CE devices still have an eBGP session established with the ISPs legacy AS, (AS64510). In some existing implementations, "Local AS No Prepend" does not concurrently modify the AS_PATH Attribute for BGP UPDATEs that are transmitted by the ISP to CE devices. Specifically, with "Local AS No Prepend" enabled on PE-B, it automatically causes a lengthening of the AS_PATH in outbound BGP UPDATEs from ISP A' toward directly attached eBGP speakers, (Customer C in AS 64496). This is the result of the "Local AS No Prepend" feature automatically appending the new global configuration ASN, AS64500, after the legacy ASN, AS64510, in BGP UPDATEs that are transmitted by PE-B to CE-B. The end result is that customer C, in AS 64496, will receive the following AS_PATH: 64510 64500 64499. Therefore, if ISP A' takes no further action, it will cause an unacceptable increase in AS_PATH length within customer's networks directly attached to ISP A'.
A second feature was designed to resolve this problem (continuing the use of Cisco CLI in the examples, it is called "Replace AS" in the examples below). This feature allows ISP A' to prevent routers configured with this feature from appending the global configured AS in outbound BGP UPDATEs toward its customer's networks configured with the "Local AS" feature. Instead, only the historical (or legacy) AS will be prepended in the outbound BGP UPDATE toward the customer's network, restoring the AS_PATH length to what it what was before AS Migration occurred.
To re-use the above diagram, but in the opposite direction, we have:
ISP A' ISP A'
CE-A ---> PE-A -------------------> PE-B ---> CE-B
64499 New_ASN: 64500 Old_ASN: 64510 64496
New_ASN: 64500
Note: Direction of BGP UPDATE as per the arrows.
Figure 4: Replace AS BGP UPDATE Diagram
The final configuration on PE-B after completing the "Replace AS" portion of the AS migration is as follows:
router bgp 64500
neighbor <CE-B_IP> remote-as 64496
neighbor <CE-B_IP> local-as 64510 no-prepend replace-as
By default, without "Replace AS" enabled, CE-B would see an AS_PATH of: 64510 64500 64499, which is artificially lengthened by the ASN Migration. After ISP A' changes PE-B to include the "Replace AS" feature, CE-B would receive an AS_PATH of: 64510 64499, which is the same AS_PATH length pre-AS migration.
While multiple implementations already exist, the following should document the expected behavior such that a new implementation of this feature could be done on other platforms.
These features MUST be configurable on a per-neighbor or per peer-group basis to allow for maximum flexibility. When this feature set is invoked, an ASN that is different from the globally-configured ASN is provided as a part of the command as exemplified above. To implement this feature, a BGP speaker MUST send BGP OPEN messages to the configured eBGP peer using the ASN configured for this session as the value sent in MY ASN. The speaker MUST NOT use the ASN configured globally within the BGP process as the value sent in MY ASN in the OPEN message. This will avoid the BGP OPEN Error message BAD PEER AS, and is typically used to re-establish eBGP sessions with peers expecting the legacy ASN after a router has been moved to a new ASN. Additionally, when the BGP speaker configured with this feature receives updates from its neighbor, it MUST process the update as normal, but it MUST append the configured ASN in the AS_PATH attribute before advertising the UPDATE to any other BGP speaker. Note that processing the update as normal will include appending the globally configured ASN to the AS_PATH, thus processing this update will result in the addition of two ASNs to the AS_PATH attribute. Similarly, for outbound updates sent by the configured BGP speaker to its neighbor, the speaker MUST append the configured ASN to the AS_PATH attribute, adding to the existing global ASN in the AS_PATH, for a total of two ASNs added to the AS_PATH.
Two options exist to manipulate the behavior of this feature. They modify the behavior as described below:
No prepend inbound – When the BGP speaker configured with this option receives inbound updates from its neighbor, it MUST NOT append the configured ASN in the AS_PATH attribute when advertising that UPDATE to other peers and instead MUST append only the globally configured ASN.
No prepend outbound – When the BGP speaker configured with this option generates outbound BGP updates to the configured peer, the BGP speaker MUST remove the globally configured ASN from the AS_PATH attribute, and MUST append the locally configured ASN to the AS_PATH attribute before sending outbound BGP updates to the configured peer.
While the exact command syntax is an implementation detail beyond the scope of this document, the following consideration may be helpful for implementers: Implementations MAY integrate the behavior of the options described above into a single command that addresses both inbound and outbound updates, but if this is done, implementations MUST provide a method to select its applicability to inbound updates, outbound updates, or updates in both directions. Several existing implementations use separate commands (e.g. local-as no-prepend vs local-as replace-as) for maximum flexibility in controlling the behavior on the session to address the widest range of possible migration scenarios.
The following section describes features that assist with a gradual and least service impacting migration of Internal BGP sessions from a legacy ASN to the permanently retained ASN. The following feature is very valuable to networks undergoing AS migration, but its use does not cause changes to the AS_PATH attribute.
In this case, all of the routers to be consolidated into a single, permanently retained ASN are under the administrative control of a single entity. Unfortunately, the traditional method of migrating all Internal BGP speakers, particularly within larger networks, is both time consuming and widely service impacting.
The traditional method to migrate Internal BGP sessions was strictly limited to reconfiguration of the global configuration ASN and, concurrently, changing all iBGP neighbors' remote ASN from the legacy ASN to the new, permanently retained ASN on each router within the legacy AS. These changes can be challenging to swiftly execute in networks with with more than a few dozen internal BGP speakers. There is also the concomitant service interruptions as these changes are made to routers within the network, resulting in a reset of iBGP sessions and subsequent route reconvergence to reestablish optimal routing paths. Operators often cannot make such sweeping changes given the associated risks of a highly visible service interruption; rather, they require a more gradual method to migrate Internal BGP sessions, from one ASN to a second, permanently retained ASN, that is not visibly service-impacting to its customers.
With the "Internal BGP Alias" [JUNIPER] feature, it allows an Internal BGP speaker to form a single iBGP session using either the old, legacy ASN or the new, permanently retained ASN. The benefits of using this feature are several fold. First, it allows for a more gradual and less service-impacting migration away from the legacy ASN to the permanently retained ASN. Second, it (temporarily) permits the coexistence of the legacy and permanently retained ASN within a single network, allowing for uniform BGP path selection among all routers within the consolidated network. NB: Cisco doesn't have an exact equivalent to "Internal BGP Alias", but the combination of the Cisco features iBGP local-AS and dual-as provides similar functionality.
When the "Internal BGP Alias" feature is enabled, typically just on one side of a iBGP session, it allows that iBGP speaker to establish a single iBGP session with either the legacy ASN or the new, permanently retained ASN, depending on which one it receives in the "My Autonomous System" field of the BGP OPEN message from its iBGP session neighbor. It is important to recognize that enablement of the "Internal BGP Alias" feature preserves the semantics of a regular iBGP session, (using identical ASNs). Thus, the BGP attributes transmitted by and the acceptable methods of operation on BGP attributes received from iBGP sessions configured with "Internal BGP Alias" are no different than those exchanged across an iBGP session without "Internal BGP Alias" configured, as defined by [RFC4271] and [RFC4456].
Typically, in medium to large networks, BGP Route Reflectors [RFC4456] (RRs) are used to aid in reduction of configuration of iBGP sessions and scalability with respect to overall TCP (and, BGP) session maintenance between adjacent iBGP speakers. Furthermore, BGP Route Reflectors are typically deployed in pairs within a single Route Reflection cluster to ensure high reliability of the BGP Control Plane. As such, the following example will use Route Reflectors to aid in understanding the use of the "Internal BGP Alias" feature. Note that Route Reflectors are not a prerequisite to enable "Internal BGP Alias" and this feature can be enabled independent of the use of Route Reflectors.
The general order of operations is as follows:
The benefit of using "Internal BGP Alias" is that it is a more gradual and less externally service-impacting change to accomplish an AS migration. Previously, without "Internal BGP Alias", such an AS migration change would carry a high risk and need to be successfully accomplished in a very short timeframe (e.g.: at most several hours). In addition, it would likely cause substantial routing churn and rapid fluctuations in traffic carried -- potentially causing periods of congestion and resultant packet loss -- during the period the configuration changes are underway to complete the AS Migration. On the other hand, with "Internal BGP Alias", the migration from the legacy ASN to the permanently retained ASN can occur over a period of days or weeks with reduced customer disruption. (The only observable service disruption should be when each PE undergoes the changes discussed in step 4 above.)
When configured with this feature, a BGP speaker MUST accept BGP OPEN and establish an iBGP session from configured iBGP peers if the ASN value in MY ASN is either the globally configured ASN or the locally configured ASN provided in this command. Additionally, a BGP speaker configured with this feature MUST send its own BGP OPEN using both the globally configured and the locally configured ASN in MY ASN. To avoid potential deadlocks when two BGP speakers are attempting to establish a BGP peering session and are both configured with this feature, the speaker SHOULD send BGP OPEN using the globally configured ASN first, and only send a BGP OPEN using the locally configured ASN as a fallback if the remote neighbor responds with the BGP error BAD PEER ASN. In each case, the BGP speaker MUST treat updates sent and received to this peer as if this was a natively configured iBGP session, as defined by [RFC4271] and [RFC4456].
Implementations of this feature MAY integrate the functionality from the eBGP features [ebgp-features] section as a part of this command in order to simplify support for eBGP migrations as well as iBGP migrations, such that an eBGP session to a configured neighbor could be established via either the global ASN or the locally configured ASN. If the eBGP session is established with the global ASN, no modifications to AS_PATH are required, but if the eBGP session is established with the locally configured ASN, the modifications discussed in eBGP features [ebgp-features] MUST be implemented to properly manipulate the AS_PATH.
This document describes several features to support ISPs and other organizations that need to perform ASN migrations. Other variations of these features may exist, for example, in legacy router software that has not been upgraded or reached End of Life, but continues to operate in the network. Such variations are beyond the scope of this document.
Companies routinely go through periods of mergers, acquisitions and divestitures, which in the case of the former cause them to accumulate several legacy ASNs over time. ISPs often do not have control over the configuration of customers' devices (i.e.: the ISPs are often not providing a managed CE router service, particularly to medium and large customers that require eBGP). Furthermore, ISPs are using methods to perform ASN migration that do not require coordination with customers. Ultimately, this means there is not a finite period of time after which legacy ASNs will be completely expunged from the ISP's network. In fact, it is common that legacy ASNs and the associated External BGP AS Migration features discussed in this document can and do persist for several years, if not longer. Thus, it is prudent to plan that legacy ASNs and associated External BGP AS Migration features will persist in a operational network indefinitely.
With respect to the Internal BGP AS Migration Features, all of the routers to be consolidated into a single, permanently retained ASN are under the administrative control of a single entity. Thus, completing the migration from iBGP sessions using the legacy ASN to the permanently retained ASN is more straightforward and could be accomplished in a matter of days to months. Finally, good operational hygiene would dictate that it is good practice to avoid using "Internal BGP Alias" over a long period of time for reasons of not only operational simplicity of the network, but also reduced reliance on that feature during the ongoing lifecycle management of software, features and configurations that are maintained on the network.
Although the features discussed in this document are not formally recognized as part of the BGP4 specification, they have been in existence in commercial implementations for well over a decade. These features are widely known by the operational community and will continue to be a critical necessity in the support of network integration activities going forward. Therefore, these features are extremely unlikely to be deprecated by vendors. As a result, these features must be acknowledged by protocol designers, particularly when there are proposals to modify BGP's behavior with respect to handling or manipulation of the AS_PATH Attribute. More specifically, assumptions should not be made with respect to the preservation or consistency of the AS_PATH Attribute as it is transmitted along a sequence of ASNs. In addition, proposals to manipulate the AS_PATH that would gratuitously increase AS_PATH length or remove the capability to use these features described in this document will not be accepted by the operational community.
Thanks to Kotikalapudi Sriram, Stephane Litkowski, Terry Manderson, David Farmer, Jaroslaw Adam Gralak, Gunter Van de Velde, Juan Alcaide, Jon Mitchell, and Thomas Morin for their comments.
This memo includes no request to IANA.
This draft discusses a process by which one ASN is migrated into and subsumed by another. This involves manipulating the AS_PATH Attribute with the intent of not increasing the AS_PATH length, which would typically cause the BGP route to no longer be selected by BGP's Path Selection Algorithm in others' networks. This could result in a loss of revenue if the ISP is billing based on measured utilization of traffic sent to/from entities attached to its network. This could also result in sudden and unexpected shifts in traffic patterns in the network, potentially resulting in congestion, in the most extreme cases.
Given that these features can only be enabled through configuration of routers within a single network, standard security measures should be taken to restrict access to the management interface(s) of routers that implement these features.
As noted elsewhere in this document, this set of migration features has multiple existing implementations in wide use.
This is not intended to be an exhaustive list, as equivalent features do exist in other implementations, however the authors were unable to find publicly available documentation of the vendor-specific implementation to reference.
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. |
| [RFC4271] | Rekhter, Y., Li, T. and S. Hares, "A Border Gateway Protocol 4 (BGP-4)", RFC 4271, January 2006. |
| [RFC5398] | Huston, G., "Autonomous System (AS) Number Reservation for Documentation Use", RFC 5398, December 2008. |
| [ALU] | Alcatel-Lucent, "BGP Local AS attribute", 2006-2012. |
| [CISCO] | Cisco Systems, Inc., "BGP Support for Dual AS Configuration for Network AS Migrations", 2003. |
| [JUNIPER] | Juniper Networks, Inc., "Configuring the BGP Local Autonomous System Attribute", 2012. |
| [RFC4456] | Bates, T., Chen, E. and R. Chandra, "BGP Route Reflection: An Alternative to Full Mesh Internal BGP (IBGP)", RFC 4456, April 2006. |
| [RFC5065] | Traina, P., McPherson, D. and J. Scudder, "Autonomous System Confederations for BGP", RFC 5065, August 2007. |
| [RFC6996] | Mitchell, J., "Autonomous System (AS) Reservation for Private Use", BCP 6, RFC 6996, July 2013. |