Network Working Group Ronald Bonica
INTERNET DRAFT MCI
Expiration Date: November 2004 Luyuan Fang
AT&T
Pedro Marques
Juniper Networks
Luca Martini
Robert Raszuk
Cisco Systems
Constrained VPN route distribution
draft-ietf-l3vpn-rt-constrain-00.txt
Status of this Memo
This document is an Internet-Draft and is in full conformance with
all provisions of Section 10 of RFC2026.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet-
Drafts.
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.''
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html.
Abstract
This document defines MP-BGP procedures that allow BGP speakers to
exchange Route Target reachability information. This information can
be used to build a route distribution graph in order to limit the
propagation of VPN NLRI (such as VPN-IPv4, VPN-IPv6 or L2-VPN NLRI)
between different autonomous systems or distinct clusters of the same
autonomous system.
draft-ietf-l3vpn-rtd-rcaofnts-tireatifn--l030v.ptnx-trt-constrain-00.txt [Page 1]
Internet Draft May 2004
1. Introduction
In BGP/MPLS IP VPNs, PE routers use Route Target (RT) extended
communities to control the distribution of routes into VRFs. Within a
given iBGP mesh, PE routers need only to hold routes marked with
Route Targets pertaining to VRFs that have local CE attachments.
It is common, however, for an autonomous system use route reflection
[BGP-RR] in order to simplify the process of bringing up a new PE
router in the network and to limit the size of the iBGP peering mesh.
In such a scenario, as well as when VPNs may have members in more
than one autonomous system, the number of routes carried by the
inter-cluster or inter-as distribution routers is an important
consideration.
In order to limit the VPN routing information that is maintained at a
given RR, RFC2547bis [RFC2547bis] suggests, in section 4.3.3., the
usage of "Cooperative Route Filtering" [ORF] between route
reflectors.
As currently defined, "Cooperative Route Filtering" has a fundamental
limitation in that it can only distribute information in a point-to-
point fashion. As such, it doesn't lend itself to be used to control
the propagation of VPN NLRI information, either in an hierarchical
way within an autonomous system, or between autonomous systems.
This limitation conditions the effectiveness of the suggestions
presented in section 4.3.3. of RFC2547bis [RFC2547bis] in terms of
their ability to limit the number of VPN routes known to the RRs. Of
these, option 2 proposes that route reflectors build their inter-
cluster Route Target filter based on the routes received from client
PE routers. This assumes a symmetric model in which a VPN uses the
same Route Target value for both Import and Export targets. An
asymmetric model, such as an hub-and-spoke scenario, would not be
supported by this suggestion. This proposal addresses this issue by
basing itself on the Import Targets that define the VPN NLRI to VRF
mapping.
While it would be possible to extend the encoding currently defined
for extended-community ORF in order to achieve this purpose, BGP
itself already has all the necessary machinery for dissemination of
arbitrary information in a loop free fashion, both within a single
autonomous system, as well as across multiple autonomous systems.
This document builds on the model described in RFC2547bis and on
concept of cooperative route filtering by adding the ability to
propagate Route Target information between iBGP meshes.
draft-ietf-l3vpn-rtd-rcaofnts-tireatifn--l030v.ptnx-trt-constrain-00.txt [Page 2]
Internet Draft May 2004
By using MP-BGP UPDATE messages to propagate Route Target information
it is possible to reuse all this machinery including route
reflection, confederations and inter-as information loop detection.
Received Route Target information can then be used to restrict
advertisement of VPN NLRI to peers that have advertised their
respective Route Targets, effectively building a route distribution
graph. In this model, VPN NLRI routing information flows in the
inverse direction of Route Target information.
This mechanism is applicable to any BGP NLRI that controls the
distribution of routing information based on Route Targets, such as
BGP L2VPNs [L2VPN] and VPLS [VPLS].
Throughout this document, the term NLRI, which originally expands to
"Network Layer Reachability Information" is used to describe routing
information carried via MP-BGP updates without any assumption of
semantics.
2. Inter-AS VPN route distribution.
In order to better understand the problem at hand, it is helpful to
divide it in its inter-AS and intra-AS components. Figure 1
represents an arbitrary graph of autonomous systems (a through j)
interconnected in an ad-hoc fashion. The following discussion
ignores the complexity of intra-AS route distribution.
+----------------------------------+
| +---+ +---+ +---+ |
| | a | -- | b | -- | c | |
| +---+ +---+ +---+ |
| | | |
| | | |
| +---+ +---+ +---+ +---+ |
| | d | -- | e | -- | f | -- | j | |
| +---+ +---+ +---+ +---+ |
| / | |
| / | |
| +---+ +---+ +---+ |
| | g | -- | h | -- | i | |
| +---+ +---+ +---+ |
+----------------------------------+
Figure 1.
Lets consider the simple case of a VPN with CE attachments in ASes a
and i using a single Route Target to control VPN route distribution.
Ideally we would like to build a flooding graph for the respective
draft-ietf-l3vpn-rtd-rcaofnts-tireatifn--l030v.ptnx-trt-constrain-00.txt [Page 3]
Internet Draft May 2004
VPN routes that would not include nodes (c, g, h, j).
In order to achieve this we will rely on ASa and ASi generating a
NLRI consisting of <as#, route-target>. Receipt of such an
advertisement by one of the ASes in the network will signal the need
to distribute VPN routes containing this Route Target community to
the peer that advertised this route.
Using routes that include both route-target and originator as#,
allows BGP speakers to use standard path selection rules concerning
as-path length (and other policy mechanisms) to prune duplicate paths
in the flooding graph, while maintaining the information required to
reach all autonomous systems advertising the Route Target.
In the example above, ASe needs to maintain a path to ASa in order to
flood VPN routing information originating from ASi and vice-versa. It
should however prune less preferred paths such as the longer path to
ASi with as-path (g h i).
Extending the example above to include ASj as a member of the VPN
distribution graph would cause ASf to advertise 2 Route Target routes
to e, one containing origin ASi and one containing origin ASj. While
advertising a single path, lets assume (f j) is selected, would be
sufficient to guarantee that VPN information flows to all VPN member
ASes, the information concerning the path (f i) is necessary to prune
the arc (g h i) from the route distribution graph.
As with other approaches for building distribution graphs, the
benefits of this mechanism are directly proportional to how "sparse"
is the VPN membership. Standard RFC2547 inter-AS behavior can be seen
as a dense-mode approach, to make the analogy with multicast routing
protocols.
3. Intra-AS VPN route distribution
As indicated above, the inter-AS VPN route distribution graph, for a
given route-target, is constructed by creating a directed arc on the
inverse direction of received Route Target UPDATEs containing an NLRI
of the form <as#, route-target>.
Inside the BGP topology of a given autonomous-system, as far as
external routes are concerned (route-targets where the as# is not the
local as), it is easy to see that standard BGP route selection and
advertisement rules [BGP-BASE] will allow a transit AS to create the
necessary flooding state.
Consider a IPv4 NLRI prefix, sourced by a single AS, which
draft-ietf-l3vpn-rtd-rcaofnts-tireatifn--l030v.ptnx-trt-constrain-00.txt [Page 4]
Internet Draft May 2004
distributed via BGP within a given transit AS. BGP protocol rules
guarantee that BGP speaker has a valid route that can be used for
forwarding of data packets for that destination prefix, in the
inverse path of received routing updates.
By the same token, and given that a <as#, route-target> key provides
uniqueness between several ASes that may be sourcing this route-
target, BGP route selection and advertisement procedures guarantee
that a valid VPN route distribution path exists to the origin of the
Route Target advertisement.
Route Target routes that are originated within the autonomous-system
however require more careful examination. Several PE routers within a
given autonomous-system may source the the same NLRI <as#, route-
target>, thus default route advertisement rules are no longer
sufficient to guarantee that within the given AS each node in the
distribution graph has selected a feasible path to each of the PEs
that import the given route-target.
When processing Route Target routes for which the as# is equal to the
local autonomous system, it is necessary to consider all availiable
iBGP paths for a given RT prefix when performing outbound route
filtering, not just the best path.
In addition, when advertising Route Target NLRI information sourced
by the local autonomous system to an iBGP peer, a BGP speaker shall
modify its procedure to calculate the BGP attributes such that:
When advertising a route to a route-reflector client, the
Originator attribute shall be set to the router-id of the
advertiser and the Next-hop attribute shall be set of the local
address for that session.
When advertising a route to a non client peer, if the best path as
selected by path selection procedure described in section 9.1 of
[BGP-BASE], is a route received from a non-client peer, and there
is an alternative path to the same destination from a client, the
attributes of the client path are advertised to the peer.
The first of these route advertisement rules is designed such that
the originator of a route does not drop a route which is reflected
back to it, thus allowing the route reflector to use this route in
order to signal the client that it should distribute VPN routes with
the specific target torwards the reflector.
The second rule makes is such that any BGP speaker present in an iBGP
mesh can signal the interest of its route reflection clients in
receiving VPN routes for that target.
draft-ietf-l3vpn-rtd-rcaofnts-tireatifn--l030v.ptnx-trt-constrain-00.txt [Page 5]
Internet Draft May 2004
An alternative solution to the procedure given above would have been
to source different routes per PE, such as NLRI of the form
<originator-id, route-target>, and aggregate them at the edge of the
network. The solution adopted is considered to be advantageous over
the former given that it requires less routing-information within a
given AS.
4. Route Target advertisements
Route Target routing information is advertised in BGP UPDATE messages
using the MP_REACH_NLRI and MP_UNREACH_NLRI attributes [BGP-MP]. The
<AFI, SAFI> value pair used to identify this NLRI is (AFI=1,
SAFI=132).
The Next Hop field of MP_REACH_NLRI attribute shall be interpreted as
an IPv4 address, whenever the lenght of NextHop address is 4 octects,
and as a IPv6 address, whenever the lenght of the NextHop address is
16 octets.
The NLRI field in the MP_REACH_NLRI and MP_UNREACH_NLRI is a prefix
of 0 to 96 bits encoded as defined in section 4 of [BGP-MP].
This prefix is structured as follows:
+-------------------------------+
| origin as (4 octects) |
+-------------------------------+
| route target (8 octects) |
+ +
| |
+-------------------------------+
Except for the default route target, which is encoded as a 0 lenght
prefix, the minimum prefix lenght is 32 bits. Thus, the origin AS
must be set on a prefix.
Route targets can then be expressed as prefixes, where for instance,
a prefix would encompass all route target extended communities
assigned by a given Global Administrator [BGP-EXTCOMM].
The default route target can be used to indicate to a peer the
willingness to receive all VPN route advertisements such as, for
instance, the case of route reflector speaking to one of its PE
router clients.
draft-ietf-l3vpn-rtd-rcaofnts-tireatifn--l030v.ptnx-trt-constrain-00.txt [Page 6]
Internet Draft May 2004
5. Capability Advertisement
A BGP speaker that wishes to exchange Route Target information must
use the the Multiprotocol Extensions Capability Code as defined in
[BGP-MP], to advertise the corresponding (AFI, SAFI) pair.
A BGP speaker MAY participate in the distribution of Route Target
information while not using the learned information for purposes of
VPN NLRI route filtering, although the latter is discouraged.
6. Operation
A VPN NLRI route should be advertised to a peer that participates in
the exchange of Route Target information if that peer has advertised
either the default Route Target or any of the targets contained in
the extended communities attribute of the VPN route in question.
When a BGP speaker receives a BGP UPDATE that advertises or withdraws
a given Route Target, it should examine the RIB-OUTs of VPN NLRIs and
reevaluate the advertisement status of routes that match the Route
Target in question.
A BGP speaker should generate the minimum set of BGP VPN route
updates necessary to transition between the previous and current
state of the route distribution graph that is derived from Route
Target information.
7. Deployment considerations
This mecanism reduces the scaling requirements that are imposed on
route reflectors by limiting the number of VPN routes and events that
a reflector has to process to the VPN routes used by its direct
clients. By default, a reflector must scale in terms of the total
number of VPN routes present on the network.
This also means that its is now possible to reduce the load impossed
on a given reflector by dividing the PE routers present on its
cluster into a new set of clusters. This is a localized configuration
change that need not affect any system outside this cluster.
The effectiveness of RT-based filtering depends on how sparse the VPN
membership is.
For instance, in the inter-as case, it is likely that a given VPN is
connected to only a subset of all participating ASes. The only
current mechanism to limit the scope of VPN route flooding is through
draft-ietf-l3vpn-rtd-rcaofnts-tireatifn--l030v.ptnx-trt-constrain-00.txt [Page 7]
Internet Draft May 2004
manual filtering on the EBGP border routers. With the current
proposal such filtering will be performed based on the dynamic RT-
route information.
In some inter-as deployments not all RTs used for a given VPN have
external significance. For example, a VPN can use an hub RT and a
spoke RT internally to an autonomous-system. The spoke RT does not
have meaning outside this AS and so it may be stripped at an external
border router. The same policy rules that result in extended
community filtering can be applied to RT-route filtering in order to
avoid advertising an RT-route for the spoke-RT in the example above.
Throughout this document, we assume that autonomous-systems agree on
an RT assignment convention. RT translation at the extern border
router boundary, is considered to be a local implementation decision,
as it should not affect inter-operability.
8. Security considerations
This document does not alter the security properties of BGP-based
VPNs.
9. Acknowledgments
This proposal is based on the extended community route filtering
mechanism defined in [ORF].
Ahmed Guetari was instrumental in defining requirements for this
proposal.
The authors would also like to thank Yakov Rekhter, Dan Tappan, Dave
Ward, John Scudder, Keyur Patel, and Jerry Ash for their comments and
suggestions.
draft-ietf-l3vpn-rtd-rcaofnts-tireatifn--l030v.ptnx-trt-constrain-00.txt [Page 8]
Internet Draft May 2004
10. References
[BGP-BASE] Y. Rekhter, T. Li, S. Hares, "A Border Gateway Protocol 4
(BGP-4)", draft-ietf-idr-bgp4-20.txt, 03/03
[RFC2547bis] "BGP/MPLS VPNs", Rosen et. al., draft-ietf-ppvpn-
rfc2547bis-03.txt, 10/02.
[BGP-RR] Bates, Chandra, and Chen, "BGP Route Reflection: An
alternative to full mesh IBGP", RFC 2796.
[BGP-CAP] R. Chandra, J. Scudder, "Capabilities Advertisement with
BGP-4", RFC2842.
[BGP-MP] T. Bates, R. Chandra, D. Katz, Y. Rekhter, "Multiprotocol
Extensions for BGP-4", RFC2858.
[ORF] E. Chen, Y. Rekhter, "Cooperative Route Filtering Capability
for BGP-4", draft-ietf-idr-route-filter-09.txt, 08/03.
[BGP-EXTCOMM] S. Sangli, D. Tappan, Y. Rekhter, "BGP Extended
Communities Attribute", draft-ietf-idr-bgp-ext-communities-05.txt,
05/02.
[L2VPN] K. Kompella et al., "Layer 2 VPNs Over Tunnels", draft-
kompella-ppvpn-l2vpn-02.txt, 11/01.
[VPLS] K Kompella (Ed.), "Virtual Private LAN Service", draft-
kompella-ppvpn-vpls-01.txt, 11/02
11. Authors' Addresses
draft-ietf-l3vpn-rtd-rcaofnts-tireatifn--l030v.ptnx-trt-constrain-00.txt [Page 9]
Internet Draft May 2004
Ronald P. Bonica
MCI
22001 Loudoun County Pkwy
Ashburn, Virginia, 20147
Phone: 703 886 1681
Email: ronald.p.bonica@mci.com
Luyuan Fang
AT&T
200 Laurel Avenue, Room C2-3B35
Middletown, NJ 07748
Phone: 732-420-1921
Email: luyuanfang@att.com
Luca Martini
Cisco Systems, Inc.
9155 East Nichols Avenue, Suite 400
Englewood, CO, 80112
e-mail: lmartini@cisco.com
Pedro Marques
Juniper Networks
1194 N. Mathilda Ave.
Sunnyvale, CA 94089
Email: roque@juniper.net
Robert Raszuk
Cisco Systems, Inc.
170 West Tasman Dr
San Jose, CA 95134
Email: rraszuk@cisco.com
