This is a purely informative rendering of an RFC that includes verified errata. This rendering may not be used as a reference.
The following 'Verified' errata have been incorporated in this document:
EID 4144
Network Working Group M. Lasserre, Ed.
Request for Comments: 4762 V. Kompella, Ed.
Category: Standards Track Alcatel-Lucent
January 2007
Virtual Private LAN Service (VPLS) Using
Label Distribution Protocol (LDP) Signaling
Status of This Memo
This document specifies an Internet standards track protocol for the
Internet community, and requests discussion and suggestions for
improvements. Please refer to the current edition of the "Internet
Official Protocol Standards" (STD 1) for the standardization state
and status of this protocol. Distribution of this memo is unlimited.
Copyright Notice
Copyright (C) The IETF Trust (2007).
IESG Note
The L2VPN Working Group produced two separate documents, RFC 4761 and
this document, that perform similar functions using different
signaling protocols. Be aware that each method is commonly referred
to as "VPLS" even though they are distinct and incompatible with one
another.
Abstract
This document describes a Virtual Private LAN Service (VPLS) solution
using pseudowires, a service previously implemented over other
tunneling technologies and known as Transparent LAN Services (TLS).
A VPLS creates an emulated LAN segment for a given set of users;
i.e., it creates a Layer 2 broadcast domain that is fully capable of
learning and forwarding on Ethernet MAC addresses and that is closed
to a given set of users. Multiple VPLS services can be supported
from a single Provider Edge (PE) node.
This document describes the control plane functions of signaling
pseudowire labels using Label Distribution Protocol (LDP), extending
RFC 4447. It is agnostic to discovery protocols. The data plane
functions of forwarding are also described, focusing in particular on
the learning of MAC addresses. The encapsulation of VPLS packets is
described by RFC 4448.
Table of Contents
1. Introduction ....................................................3
2. Terminology .....................................................3
2.1. Conventions ................................................4
3. Acronyms ........................................................4
4. Topological Model for VPLS ......................................5
4.1. Flooding and Forwarding ....................................6
4.2. Address Learning ...........................................6
4.3. Tunnel Topology ............................................7
4.4. Loop free VPLS .............................................7
5. Discovery .......................................................7
6. Control Plane ...................................................7
6.1. LDP-Based Signaling of Demultiplexers ......................8
6.1.1. Using the Generalized PWid FEC Element ..............8
6.2. MAC Address Withdrawal .....................................9
6.2.1. MAC List TLV ........................................9
6.2.2. Address Withdraw Message Containing MAC List TLV ...11
7. Data Forwarding on an Ethernet PW ..............................11
7.1. VPLS Encapsulation Actions ................................11
7.2. VPLS Learning Actions .....................................12
8. Data Forwarding on an Ethernet VLAN PW .........................13
8.1. VPLS Encapsulation Actions ................................13
9. Operation of a VPLS ............................................14
9.1. MAC Address Aging .........................................15
10. A Hierarchical VPLS Model .....................................16
10.1. Hierarchical Connectivity ................................16
10.1.1. Spoke Connectivity for Bridging-Capable Devices ...17
10.1.2. Advantages of Spoke Connectivity ..................18
10.1.3. Spoke Connectivity for Non-Bridging Devices .......19
10.2. Redundant Spoke Connections ..............................21
10.2.1. Dual-Homed MTU-s ..................................21
10.2.2. Failure Detection and Recovery ....................22
10.3. Multi-domain VPLS Service ................................23
11. Hierarchical VPLS Model Using Ethernet Access Network .........23
11.1. Scalability ..............................................24
11.2. Dual Homing and Failure Recovery .........................24
12. Contributors ..................................................25
13. Acknowledgements ..............................................25
14. Security Considerations .......................................26
15. IANA Considerations ...........................................26
16. References ....................................................27
16.1. Normative References .....................................27
16.2. Informative References ...................................27
Appendix A. VPLS Signaling using the PWid FEC Element .............29
1. Introduction
Ethernet has become the predominant technology for Local Area Network
(LAN) connectivity and is gaining acceptance as an access technology,
specifically in Metropolitan and Wide Area Networks (MAN and WAN,
respectively). The primary motivation behind Virtual Private LAN
Services (VPLS) is to provide connectivity between geographically
dispersed customer sites across MANs and WANs, as if they were
connected using a LAN. The intended application for the end-user can
be divided into the following two categories:
- Connectivity between customer routers: LAN routing application
- Connectivity between customer Ethernet switches: LAN switching
application
Broadcast and multicast services are available over traditional LANs.
Sites that belong to the same broadcast domain and that are connected
via an MPLS network expect broadcast, multicast, and unicast traffic
to be forwarded to the proper location(s). This requires MAC address
learning/aging on a per-pseudowire basis, and packet replication
across pseudowires for multicast/broadcast traffic and for flooding
of unknown unicast destination traffic.
[RFC4448] defines how to carry Layer 2 (L2) frames over point-to-
point pseudowires (PW). This document describes extensions to
[RFC4447] for transporting Ethernet/802.3 and VLAN [802.1Q] traffic
across multiple sites that belong to the same L2 broadcast domain or
VPLS. Note that the same model can be applied to other 802.1
technologies. It describes a simple and scalable way to offer
Virtual LAN services, including the appropriate flooding of
broadcast, multicast, and unknown unicast destination traffic over
MPLS, without the need for address resolution servers or other
external servers, as discussed in [L2VPN-REQ].
The following discussion applies to devices that are VPLS capable and
have a means of tunneling labeled packets amongst each other. The
resulting set of interconnected devices forms a private MPLS VPN.
2. Terminology
Q-in-Q 802.1ad Provider Bridge extensions also known
as stackable VLANs or Q-in-Q.
Qualified learning Learning mode in which each customer VLAN is
mapped to its own VPLS instance.
Service delimiter Information used to identify a specific customer
service instance. This is typically encoded in
the encapsulation header of customer frames
(e.g., VLAN Id).
Tagged frame Frame with an 802.1Q VLAN identifier.
Unqualified learning Learning mode where all the VLANs of a single
customer are mapped to a single VPLS.
Untagged frame Frame without an 802.1Q VLAN identifier.
2.1. Conventions
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].
3. Acronyms
AC Attachment Circuit
BPDU Bridge Protocol Data Unit
CE Customer Edge device
FEC Forwarding Equivalence Class
FIB Forwarding Information Base
GRE Generic Routing Encapsulation
IPsec IP security
L2TP Layer Two Tunneling Protocol
LAN Local Area Network
LDP Label Distribution Protocol
MTU-s Multi-Tenant Unit switch
PE Provider Edge device
PW Pseudowire
STP Spanning Tree Protocol
VLAN Virtual LAN
VLAN tag VLAN Identifier
4. Topological Model for VPLS
An interface participating in a VPLS must be able to flood, forward,
and filter Ethernet frames. Figure 1, below, shows the topological
model of a VPLS. The set of PE devices interconnected via PWs
appears as a single emulated LAN to customer X. Each PE will form
remote MAC address to PW associations and associate directly attached
MAC addresses to local customer facing ports. This is modeled on
standard IEEE 802.1 MAC address learning.
+-----+ +-----+
| CE1 +---+ ........................... +---| CE2 |
+-----+ | . . | +-----+
Site 1 | +----+ +----+ | Site 2
+---| PE | Cloud | PE |---+
+----+ +----+
. .
. +----+ .
..........| PE |...........
+----+ ^
| |
| +-- Emulated LAN
+-----+
| CE3 |
+-----+
Site 3
Figure 1: Topological Model of a VPLS for
Customer X with three sites
We note here again that while this document shows specific examples
using MPLS transport tunnels, other tunnels that can be used by PWs
(as mentioned in [RFC4447]) -- e.g., GRE, L2TP, IPsec -- can also be
used, as long as the originating PE can be identified, since this is
used in the MAC learning process.
The scope of the VPLS lies within the PEs in the service provider
network, highlighting the fact that apart from customer service
delineation, the form of access to a customer site is not relevant to
the VPLS [L2VPN-REQ]. In other words, the attachment circuit (AC)
connected to the customer could be a physical Ethernet port, a
logical (tagged) Ethernet port, an ATM PVC carrying Ethernet frames,
etc., or even an Ethernet PW.
The PE is typically an edge router capable of running the LDP
signaling protocol and/or routing protocols to set up PWs. In
addition, it is capable of setting up transport tunnels to other PEs
and delivering traffic over PWs.
4.1. Flooding and Forwarding
One of attributes of an Ethernet service is that frames sent to
broadcast addresses and to unknown destination MAC addresses are
flooded to all ports. To achieve flooding within the service
provider network, all unknown unicast, broadcast and multicast frames
are flooded over the corresponding PWs to all PE nodes participating
in the VPLS, as well as to all ACs.
Note that multicast frames are a special case and do not necessarily
have to be sent to all VPN members. For simplicity, the default
approach of broadcasting multicast frames is used.
To forward a frame, a PE MUST be able to associate a destination MAC
address with a PW. It is unreasonable and perhaps impossible to
require that PEs statically configure an association of every
possible destination MAC address with a PW. Therefore, VPLS-capable
PEs SHOULD have the capability to dynamically learn MAC addresses on
both ACs and PWs and to forward and replicate packets across both ACs
and PWs.
4.2. Address Learning
Unlike BGP VPNs [RFC4364], reachability information is not advertised
and distributed via a control plane. Reachability is obtained by
standard learning bridge functions in the data plane.
When a packet arrives on a PW, if the source MAC address is unknown,
it needs to be associated with the PW, so that outbound packets to
that MAC address can be delivered over the associated PW. Likewise,
when a packet arrives on an AC, if the source MAC address is unknown,
it needs to be associated with the AC, so that outbound packets to
that MAC address can be delivered over the associated AC.
Standard learning, filtering, and forwarding actions, as defined in
[802.1D-ORIG], [802.1D-REV], and [802.1Q], are required when a PW or
AC state changes.
4.3. Tunnel Topology
PE routers are assumed to have the capability to establish transport
tunnels. Tunnels are set up between PEs to aggregate traffic. PWs
are signaled to demultiplex encapsulated Ethernet frames from
multiple VPLS instances that traverse the transport tunnels.
In an Ethernet L2VPN, it becomes the responsibility of the service
provider to create the loop-free topology. For the sake of
simplicity, we define that the topology of a VPLS is a full mesh of
PWs.
4.4. Loop free VPLS
If the topology of the VPLS is not restricted to a full mesh, then it
may be that for two PEs not directly connected via PWs, they would
have to use an intermediary PE to relay packets. This topology would
require the use of some loop-breaking protocol, like a spanning tree
protocol.
Instead, a full mesh of PWs is established between PEs. Since every
PE is now directly connected to every other PE in the VPLS via a PW,
there is no longer any need to relay packets, and we can instantiate
a simpler loop-breaking rule: the "split horizon" rule, whereby a PE
MUST NOT forward traffic from one PW to another in the same VPLS
mesh.
Note that customers are allowed to run a Spanning Tree Protocol (STP)
(e.g., as defined in [802.1D-REV]), such as when a customer has "back
door" links used to provide redundancy in the case of a failure
within the VPLS. In such a case, STP Bridge PDUs (BPDUs) are simply
tunneled through the provider cloud.
5. Discovery
The capability to manually configure the addresses of the remote PEs
is REQUIRED. However, the use of manual configuration is not
necessary if an auto-discovery procedure is used. A number of auto-
discovery procedures are compatible with this document
([RADIUS-DISC], [BGP-DISC]).
6. Control Plane
This document describes the control plane functions of signaling of
PW labels. Some foundational work in the area of support for multi-
homing is laid. The extensions to provide multi-homing support
should work independently of the basic VPLS operation, and they are
not described here.
6.1. LDP-Based Signaling of Demultiplexers
A full mesh of LDP sessions is used to establish the mesh of PWs.
The requirement for a full mesh of PWs may result in a large number
of targeted LDP sessions. Section 10 discusses the option of setting
up hierarchical topologies in order to minimize the size of the VPLS
full mesh.
Once an LDP session has been formed between two PEs, all PWs between
these two PEs are signaled over this session.
In [RFC4447], two types of FECs are described: the PWid FEC Element
(FEC type 128) and the Generalized PWid FEC Element (FEC type 129).
The original FEC element used for VPLS was compatible with the PWid
FEC Element. The text for signaling using the PWid FEC Element has
been moved to Appendix A. What we describe below replaces that with
a more generalized L2VPN descriptor, the Generalized PWid FEC
Element.
6.1.1. Using the Generalized PWid FEC Element
[RFC4447] describes a generalized FEC structure that is be used for
VPLS signaling in the following manner. We describe the assignment
of the Generalized PWid FEC Element fields in the context of VPLS
signaling.
Control bit (C): This bit is used to signal the use of the control
word as specified in [RFC4447].
PW type: The allowed PW types are Ethernet (0x0005) and Ethernet
tagged mode (0x004), as specified in [RFC4446].
PW info length: As specified in [RFC4447].
Attachment Group Identifier (AGI), Length, Value: The unique name of
this VPLS. The AGI identifies a type of name, and Length denotes the
length of Value, which is the name of the VPLS. We use the term AGI
interchangeably with VPLS identifier.
Target Attachment Individual Identifier (TAII), Source Attachment
Individual Identifier (SAII): These are null because the mesh of PWs
in a VPLS terminates on MAC learning tables, rather than on
individual attachment circuits. The use of non-null TAII and SAII is
reserved for future enhancements.
Interface Parameters: The relevant interface parameters are:
- MTU: The MTU (Maximum Transmission Unit) of the VPLS MUST be the
same across all the PWs in the mesh.
- Optional Description String: Same as [RFC4447].
- Requested VLAN ID: If the PW type is Ethernet tagged mode, this
parameter may be used to signal the insertion of the appropriate
VLAN ID, as defined in [RFC4448].
6.2. MAC Address Withdrawal
It MAY be desirable to remove or unlearn MAC addresses that have been
dynamically learned for faster convergence. This is accomplished by
sending an LDP Address Withdraw Message with the list of MAC
addresses to be removed to all other PEs over the corresponding LDP
sessions.
We introduce an optional MAC List TLV in LDP to specify a list of MAC
addresses that can be removed or unlearned using the LDP Address
Withdraw Message.
The Address Withdraw message with MAC List TLVs MAY be supported in
order to expedite removal of MAC addresses as the result of a
topology change (e.g., failure of the primary link for a dual-homed
VPLS-capable switch).
In order to minimize the impact on LDP convergence time, when the MAC
list TLV contains a large number of MAC addresses, it may be
preferable to send a MAC address withdrawal message with an empty
list.
6.2.1. MAC List TLV
MAC addresses to be unlearned can be signaled using an LDP Address
Withdraw Message that contains a new TLV, the MAC List TLV. Its
format is described below. The encoding of a MAC List TLV address is
the 6-octet MAC address specified by IEEE 802 documents [802.1D-ORIG]
[802.1D-REV].
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|U|F| Type | Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| MAC address #1 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| MAC address #1 | MAC Address #2 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| MAC address #2 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
~ ... ~
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| MAC address #n |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| MAC address #n |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
U bit: Unknown bit. This bit MUST be set to 1. If the MAC address
format is not understood, then the TLV is not understood and MUST be
ignored.
F bit: Forward bit. This bit MUST be set to 0. Since the LDP
mechanism used here is targeted, the TLV MUST NOT be forwarded.
Type: Type field. This field MUST be set to 0x0404. This identifies
the TLV type as MAC List TLV.
Length: Length field. This field specifies the total length in
octets of the MAC addresses in the TLV. The length MUST be a
multiple of 6.
MAC Address: The MAC address(es) being removed.
The MAC Address Withdraw Message contains a FEC TLV (to identify the
VPLS affected), a MAC Address TLV, and optional parameters. No
optional parameters have been defined for the MAC Address Withdraw
signaling. Note that if a PE receives a MAC Address Withdraw Message
and does not understand it, it MUST ignore the message. In this
case, instead of flushing its MAC address table, it will continue to
use stale information, unless:
- it receives a packet with a known MAC address association, but
from a different PW, in which case it replaces the old
association; or
- it ages out the old association.
The MAC Address Withdraw message only helps speed up convergence, so
PEs that do not understand the message can continue to participate in
the VPLS.
6.2.2. Address Withdraw Message Containing MAC List TLV
The processing for MAC List TLV received in an Address Withdraw
Message is:
For each MAC address in the TLV:
- Remove the association between the MAC address and the AC or PW
over which this message is received.
For a MAC Address Withdraw message with empty list:
- Remove all the MAC addresses associated with the VPLS instance
(specified by the FEC TLV) except the MAC addresses learned over
the PW associated with this signaling session over which the
message was received.
The scope of a MAC List TLV is the VPLS specified in the FEC TLV in
the MAC Address Withdraw Message. The number of MAC addresses can be
deduced from the length field in the TLV.
7. Data Forwarding on an Ethernet PW
This section describes the data plane behavior on an Ethernet PW used
in a VPLS. While the encapsulation is similar to that described in
[RFC4448], the functions of stripping the service-delimiting tag and
using a "normalized" Ethernet frame are described.
7.1. VPLS Encapsulation Actions
In a VPLS, a customer Ethernet frame without preamble is encapsulated
with a header as defined in [RFC4448]. A customer Ethernet frame is
defined as follows:
- If the frame, as it arrives at the PE, has an encapsulation that
is used by the local PE as a service delimiter, i.e., to identify
the customer and/or the particular service of that customer, then
that encapsulation may be stripped before the frame is sent into
the VPLS. As the frame exits the VPLS, the frame may have a
service-delimiting encapsulation inserted.
- If the frame, as it arrives at the PE, has an encapsulation that
is not service delimiting, then it is a customer frame whose
encapsulation should not be modified by the VPLS. This covers,
for example, a frame that carries customer-specific VLAN tags that
the service provider neither knows about nor wants to modify.
As an application of these rules, a customer frame may arrive at a
customer-facing port with a VLAN tag that identifies the customer's
VPLS instance. That tag would be stripped before it is encapsulated
in the VPLS. At egress, the frame may be tagged again, if a
service-delimiting tag is used, or it may be untagged if none is
used.
Likewise, if a customer frame arrives at a customer-facing port over
an ATM or Frame Relay VC that identifies the customer's VPLS
instance, then the ATM or FR encapsulation is removed before the
frame is passed into the VPLS.
Contrariwise, if a customer frame arrives at a customer-facing port
with a VLAN tag that identifies a VLAN domain in the customer L2
network, then the tag is not modified or stripped, as it belongs with
the rest of the customer frame.
By following the above rules, the Ethernet frame that traverses a
VPLS is always a customer Ethernet frame. Note that the two actions,
at ingress and egress, of dealing with service delimiters are local
actions that neither PE has to signal to the other. They allow, for
example, a mix-and-match of VLAN tagged and untagged services at
either end, and they do not carry across a VPLS a VLAN tag that has
local significance only. The service delimiter may be an MPLS label
also, whereby an Ethernet PW given by [RFC4448] can serve as the
access side connection into a PE. An RFC1483 Bridged PVC
encapsulation could also serve as a service delimiter. By limiting
the scope of locally significant encapsulations to the edge,
hierarchical VPLS models can be developed that provide the capability
to network-engineer scalable VPLS deployments, as described below.
7.2. VPLS Learning Actions
Learning is done based on the customer Ethernet frame as defined
above. The Forwarding Information Base (FIB) keeps track of the
mapping of customer Ethernet frame addressing and the appropriate PW
to use. We define two modes of learning: qualified and unqualified
learning. Qualified learning is the default mode and MUST be
supported. Support of unqualified learning is OPTIONAL.
In unqualified learning, all the VLANs of a single customer are
handled by a single VPLS, which means they all share a single
broadcast domain and a single MAC address space. This means that MAC
addresses need to be unique and non-overlapping among customer VLANs,
or else they cannot be differentiated within the VPLS instance, and
this can result in loss of customer frames. An application of
unqualified learning is port-based VPLS service for a given customer
(e.g., customer with non-multiplexed AC where all the traffic on a
physical port, which may include multiple customer VLANs, is mapped
to a single VPLS instance).
In qualified learning, each customer VLAN is assigned to its own VPLS
instance, which means each customer VLAN has its own broadcast domain
and MAC address space. Therefore, in qualified learning, MAC
addresses among customer VLANs may overlap with each other, but they
will be handled correctly since each customer VLAN has its own FIB;
i.e., each customer VLAN has its own MAC address space. Since VPLS
broadcasts multicast frames by default, qualified learning offers the
advantage of limiting the broadcast scope to a given customer VLAN.
Qualified learning can result in large FIB table sizes, because the
logical MAC address is now a VLAN tag + MAC address.
For STP to work in qualified learning mode, a VPLS PE must be able to
forward STP BPDUs over the proper VPLS instance. In a hierarchical
VPLS case (see details in Section 10), service delimiting tags
(Q-in-Q or [RFC4448]) can be added such that PEs can unambiguously
identify all customer traffic, including STP BPDUs. In a basic VPLS
case, upstream switches must insert such service delimiting tags.
When an access port is shared among multiple customers, a reserved
VLAN per customer domain must be used to carry STP traffic. The STP
frames are encapsulated with a unique provider tag per customer (as
the regular customer traffic), and a PEs looks up the provider tag to
send such frames across the proper VPLS instance.
8. Data Forwarding on an Ethernet VLAN PW
This section describes the data plane behavior on an Ethernet VLAN PW
in a VPLS. While the encapsulation is similar to that described in
[RFC4448], the functions of imposing tags and using a "normalized"
Ethernet frame are described. The learning behavior is the same as
for Ethernet PWs.
8.1. VPLS Encapsulation Actions
In a VPLS, a customer Ethernet frame without preamble is encapsulated
with a header as defined in [RFC4448]. A customer Ethernet frame is
defined as follows:
- If the frame, as it arrives at the PE, has an encapsulation that
is part of the customer frame and is also used by the local PE as
a service delimiter, i.e., to identify the customer and/or the
particular service of that customer, then that encapsulation is
preserved as the frame is sent into the VPLS, unless the Requested
VLAN ID optional parameter was signaled. In that case, the VLAN
tag is overwritten before the frame is sent out on the PW.
- If the frame, as it arrives at the PE, has an encapsulation that
does not have the required VLAN tag, a null tag is imposed if the
Requested VLAN ID optional parameter was not signaled.
As an application of these rules, a customer frame may arrive at a
customer-facing port with a VLAN tag that identifies the customer's
VPLS instance and also identifies a customer VLAN. That tag would be
preserved as it is encapsulated in the VPLS.
The Ethernet VLAN PW provides a simple way to preserve customer
802.1p bits.
A VPLS MAY have both Ethernet and Ethernet VLAN PWs. However, if a
PE is not able to support both PWs simultaneously, it SHOULD send a
Label Release on the PW messages that it cannot support with a status
code "Unknown FEC" as given in [RFC3036].
9. Operation of a VPLS
We show here, in Figure 2, below, an example of how a VPLS works.
The following discussion uses the figure below, where a VPLS has been
set up between PE1, PE2, and PE3. The VPLS connects a customer with
4 sites labeled A1, A2, A3, and A4 through CE1, CE2, CE3, and CE4,
respectively.
Initially, the VPLS is set up so that PE1, PE2, and PE3 have a full
mesh of Ethernet PWs. The VPLS instance is assigned an identifier
(AGI). For the above example, say PE1 signals PW label 102 to PE2
and 103 to PE3, and PE2 signals PW label 201 to PE1 and 203 to PE3.
-----
/ A1 \
---- ----CE1 |
/ \ -------- ------- / | |
| A2 CE2- / \ / PE1 \ /
\ / \ / \---/ \ -----
---- ---PE2 |
| Service Provider Network |
\ / \ /
----- PE3 / \ /
|Agg|_/ -------- -------
-| |
---- / ----- ----
/ \/ \ / \ CE = Customer Edge Router
| A3 CE3 -CE4 A4 | PE = Provider Edge Router
\ / \ / Agg = Layer 2 Aggregation
---- ----
Figure 2: Example of a VPLS
Assume a packet from A1 is bound for A2. When it leaves CE1, say it
has a source MAC address of M1 and a destination MAC of M2. If PE1
does not know where M2 is, it will flood the packet; i.e., send it to
PE2 and PE3. When PE2 receives the packet, it will have a PW label
of 201. PE2 can conclude that the source MAC address M1 is behind
PE1, since it distributed the label 201 to PE1. It can therefore
associate MAC address M1 with PW label 102.
9.1. MAC Address Aging
PEs that learn remote MAC addresses SHOULD have an aging mechanism to
remove unused entries associated with a PW label. This is important
both for conservation of memory and for administrative purposes. For
example, if a customer site A, is shut down, eventually the other PEs
should unlearn A's MAC address.
The aging timer for MAC address M SHOULD be reset when a packet with
source MAC address M is received.
10. A Hierarchical VPLS Model
The solution described above requires a full mesh of tunnel LSPs
between all the PE routers that participate in the VPLS service. For
each VPLS service, n*(n-1)/2 PWs must be set up between the PE
routers. While this creates signaling overhead, the real detriment
to large scale deployment is the packet replication requirements for
each provisioned PWs on a PE router. Hierarchical connectivity,
described in this document, reduces signaling and replication
overhead to allow large-scale deployment.
In many cases, service providers place smaller edge devices in
multi-tenant buildings and aggregate them into a PE in a large
Central Office (CO) facility. In some instances, standard IEEE
802.1q (Dot 1Q) tagging techniques may be used to facilitate mapping
CE interfaces to VPLS access circuits at a PE.
It is often beneficial to extend the VPLS service tunneling
techniques into the access switch domain. This can be accomplished
by treating the access device as a PE and provisioning PWs between it
and every other edge, as a basic VPLS. An alternative is to utilize
[RFC4448] PWs or Q-in-Q logical interfaces between the access device
and selected VPLS enabled PE routers. Q-in-Q encapsulation is
another form of L2 tunneling technique, which can be used in
conjunction with MPLS signaling, as will be described later. The
following two sections focus on this alternative approach. The VPLS
core PWs (hub) are augmented with access PWs (spoke) to form a two-
tier hierarchical VPLS (H-VPLS).
Spoke PWs may be implemented using any L2 tunneling mechanism, and by
expanding the scope of the first tier to include non-bridging VPLS PE
routers. The non-bridging PE router would extend a spoke PW from a
Layer-2 switch that connects to it, through the service core network,
to a bridging VPLS PE router supporting hub PWs. We also describe
how VPLS-challenged nodes and low-end CEs without MPLS capabilities
may participate in a hierarchical VPLS.
For rest of this discussion we refer to a bridging capable access
device as MTU-s and a non-bridging capable PE as PE-r. We refer to a
routing and bridging capable device as PE-rs.
10.1. Hierarchical Connectivity
This section describes the hub and spoke connectivity model and
describes the requirements of the bridging capable and non-bridging
MTU-s devices for supporting the spoke connections.
10.1.1. Spoke Connectivity for Bridging-Capable Devices
In Figure 3, below, three customer sites are connected to an MTU-s
through CE-1, CE-2, and CE-3. The MTU-s has a single connection
(PW-1) to PE1-rs. The PE-rs devices are connected in a basic VPLS
full mesh. For each VPLS service, a single spoke PW is set up
between the MTU-s and the PE-rs based on [RFC4447]. Unlike
traditional PWs that terminate on a physical (or a VLAN-tagged
logical) port, a spoke PW terminates on a virtual switch instance
(VSI; see [L2FRAME]) on the MTU-s and the PE-rs devices.
PE2-rs
+--------+
| |
| -- |
| / \ |
CE-1 | \S / |
\ | -- |
\ +--------+
\ MTU-s PE1-rs / |
+--------+ +--------+ / |
| | | | / |
| -- | PW-1 | -- |---/ |
| / \--|- - - - - - - - - - - | / \ | |
| \S / | | \S / | |
| -- | | -- |---\ |
+--------+ +--------+ \ |
/ \ |
---- +--------+
|Agg | | |
---- | -- |
/ \ | / \ |
CE-2 CE-3 | \S / |
| -- |
+--------+
PE3-rs
Agg = Layer-2 Aggregation
--
/ \
\S / = Virtual Switch Instance
--
Figure 3: An example of a hierarchical VPLS model
The MTU-s and the PE-rs treat each spoke connection like an AC of the
VPLS service. The PW label is used to associate the traffic from the
spoke to a VPLS instance.
10.1.1.1. MTU-s Operation
An MTU-s is defined as a device that supports layer-2 switching
functionality and does all the normal bridging functions of learning
and replication on all its ports, including the spoke, which is
treated as a virtual port. Packets to unknown destinations are
replicated to all ports in the service including the spoke. Once the
MAC address is learned, traffic between CE1 and CE2 will be switched
locally by the MTU-s, saving the capacity of the spoke to the PE-rs.
Similarly traffic between CE1 or CE2 and any remote destination is
switched directly onto the spoke and sent to the PE-rs over the
point-to-point PW.
Since the MTU-s is bridging capable, only a single PW is required per
VPLS instance for any number of access connections in the same VPLS
service. This further reduces the signaling overhead between the
MTU-s and PE-rs.
If the MTU-s is directly connected to the PE-rs, other encapsulation
techniques, such as Q-in-Q, can be used for the spoke.
10.1.1.2. PE-rs Operation
A PE-rs is a device that supports all the bridging functions for VPLS
service and supports the routing and MPLS encapsulation; i.e., it
supports all the functions described for a basic VPLS, as described
above.
The operation of PE-rs is independent of the type of device at the
other end of the spoke. Thus, the spoke from the MTU-s is treated as
a virtual port, and the PE-rs will switch traffic between the spoke
PW, hub PWs, and ACs once it has learned the MAC addresses.
10.1.2. Advantages of Spoke Connectivity
Spoke connectivity offers several scaling and operational advantages
for creating large-scale VPLS implementations, while retaining the
ability to offer all the functionality of the VPLS service.
- Eliminates the need for a full mesh of tunnels and full mesh of
PWs per service between all devices participating in the VPLS
service.
- Minimizes signaling overhead, since fewer PWs are required for the
VPLS service.
- Segments VPLS nodal discovery. MTU-s needs to be aware of only
the PE-rs node, although it is participating in the VPLS service
that spans multiple devices. On the other hand, every VPLS PE-rs
must be aware of every other VPLS PE-rs and all of its locally
connected MTU-s and PE-r devices.
- Addition of other sites requires configuration of the new MTU-s
but does not require any provisioning of the existing MTU-s
devices on that service.
- Hierarchical connections can be used to create VPLS service that
spans multiple service provider domains. This is explained in a
later section.
Note that as more devices participate in the VPLS, there are more
devices that require the capability for learning and replication.
10.1.3. Spoke Connectivity for Non-Bridging Devices
In some cases, a bridging PE-rs may not be deployed, or a PE-r might
already have been deployed. In this section, we explain how a PE-r
that does not support any of the VPLS bridging functionality can
participate in the VPLS service.
In Figure 4, three customer sites are connected through CE-1, CE-2,
and CE-3 to the VPLS through PE-r. For every attachment circuit that
participates in the VPLS service, PE-r creates a point-to-point PW
that terminates on the VSI of PE1-rs.
PE2-rs
+--------+
| |
| -- |
| / \ |
CE-1 | \S / |
\ | -- |
\ +--------+
\ PE-r PE1-rs / |
+--------+ +--------+ / |
|\ | | | / |
| \ | PW-1 | -- |---/ |
| ------|- - - - - - - - - - - | / \ | |
| -----|- - - - - - - - - - - | \S / | |
| / | | -- |---\ |
+--------+ +--------+ \ |
/ \ |
---- +--------+
| Agg| | |
---- | -- |
/ \ | / \ |
CE-2 CE-3 | \S / |
| -- |
+--------+
PE3-rs
Figure 4: An example of a hierarchical VPLS
with non-bridging spokes
The PE-r is defined as a device that supports routing but does not
support any bridging functions. However, it is capable of setting up
PWs between itself and the PE-rs. For every port that is supported
in the VPLS service, a PW is set up from the PE-r to the PE-rs. Once
the PWs are set up, there is no learning or replication function
required on the part of the PE-r. All traffic received on any of the
ACs is transmitted on the PW. Similarly, all traffic received on a
PW is transmitted to the AC where the PW terminates. Thus, traffic
from CE1 destined for CE2 is switched at PE1-rs and not at PE-r.
Note that in the case where PE-r devices use Provider VLANs (P-VLAN)
as demultiplexers instead of PWs, PE1-rs can treat them as such and
map these "circuits" into a VPLS domain to provide bridging support
between them.
This approach adds more overhead than the bridging-capable (MTU-s)
spoke approach, since a PW is required for every AC that participates
in the service versus a single PW required per service (regardless of
ACs) when an MTU-s is used. However, this approach offers the
advantage of offering a VPLS service in conjunction with a routed
internet service without requiring the addition of new MTU-s.
10.2. Redundant Spoke Connections
An obvious weakness of the hub and spoke approach described thus far
is that the MTU-s has a single connection to the PE-rs. In case of
failure of the connection or the PE-rs, the MTU-s suffers total loss
of connectivity.
In this section, we describe how the redundant connections can be
provided to avoid total loss of connectivity from the MTU-s. The
mechanism described is identical for both, MTU-s and PE-r devices.
10.2.1. Dual-Homed MTU-s
To protect from connection failure of the PW or the failure of the
PE-rs, the MTU-s or the PE-r is dual-homed into two PE-rs devices.
The PE-rs devices must be part of the same VPLS service instance.
In Figure 5, two customer sites are connected through CE-1 and CE-2
to an MTU-s. The MTU-s sets up two PWs (one each to PE1-rs and
PE3-rs) for each VPLS instance. One of the two PWs is designated as
primary and is the one that is actively used under normal conditions,
whereas the second PW is designated as secondary and is held in a
standby state. The MTU-s negotiates the PW labels for both the
primary and secondary PWs, but does not use the secondary PW unless
the primary PW fails. How a spoke is designated primary or secondary
is outside the scope of this document. For example, a spanning tree
instance running between only the MTU-s and the two PE-rs nodes is
one possible method. Another method could be configuration.
PE2-rs
+--------+
| |
| -- |
| / \ |
CE-1 | \S / |
\ | -- |
\ +--------+
\ MTU-s PE1-rs / |
+--------+ +--------+ / |
| | | | / |
| -- | Primary PW | -- |---/ |
| / \ |- - - - - - - - - - - | / \ | |
| \S / | | \S / | |
| -- | | -- |---\ |
+--------+ +--------+ \ |
/ \ \ |
/ \ +--------+
/ \ | |
CE-2 \ | -- |
\ Secondary PW | / \ |
- - - - - - - - - - - - - - - - - | \S / |
| -- |
+--------+
PE3-rs
Figure 5: An example of a dual-homed MTU-s
10.2.2. Failure Detection and Recovery
The MTU-s should control the usage of the spokes to the PE-rs
devices. If the spokes are PWs, then LDP signaling is used to
negotiate the PW labels, and the hello messages used for the LDP
session could be used to detect failure of the primary PW. The use
of other mechanisms that could provide faster detection failures is
outside the scope of this document.
Upon failure of the primary PW, MTU-s immediately switches to the
secondary PW. At this point, the PE3-rs that terminates the
secondary PW starts learning MAC addresses on the spoke PW. All
other PE-rs nodes in the network think that CE-1 and CE-2 are behind
PE1-rs and may continue to send traffic to PE1-rs until they learn
that the devices are now behind PE3-rs. The unlearning process can
take a long time and may adversely affect the connectivity of
higher-level protocols from CE1 and CE2. To enable faster
convergence, the PE3-rs where the secondary PW got activated may send
out a flush message (as explained in Section 6.2), using the MAC List
TLV, as defined in Section 6, to all PE-rs nodes. Upon receiving the
message, PE-rs nodes flush the MAC addresses associated with that
VPLS instance.
10.3. Multi-domain VPLS Service
Hierarchy can also be used to create a large-scale VPLS service
within a single domain or a service that spans multiple domains
without requiring full mesh connectivity between all VPLS-capable
devices. Two fully meshed VPLS networks are connected together using
a single LSP tunnel between the VPLS "border" devices. A single
spoke PW per VPLS service is set up to connect the two domains
together.
When more than two domains need to be connected, a full mesh of
inter-domain spokes is created between border PEs. Forwarding rules
over this mesh are identical to the rules defined in Section 4.
This creates a three-tier hierarchical model that consists of a hub-
and-spoke topology between MTU-s and PE-rs devices, a full-mesh
topology between PE-rs, and a full mesh of inter-domain spokes
between border PE-rs devices.
This document does not specify how redundant border PEs per domain
per VPLS instance can be supported.
11. Hierarchical VPLS Model Using Ethernet Access Network
In this section, the hierarchical model is expanded to include an
Ethernet access network. This model retains the hierarchical
architecture discussed previously in that it leverages the full-mesh
topology among PE-rs devices; however, no restriction is imposed on
the topology of the Ethernet access network (e.g., the topology
between MTU-s and PE-rs devices is not restricted to hub and spoke).
The motivation for an Ethernet access network is that Ethernet-based
networks are currently deployed by some service providers to offer
VPLS services to their customers. Therefore, it is important to
provide a mechanism that allows these networks to integrate with an
IP or MPLS core to provide scalable VPLS services.
One approach of tunneling a customer's Ethernet traffic via an
Ethernet access network is to add an additional VLAN tag to the
customer's data (which may be either tagged or untagged). The
additional tag is referred to as Provider's VLAN (P-VLAN). Inside
the provider's network each P-VLAN designates a customer or more
specifically a VPLS instance for that customer. Therefore, there is
a one-to-one correspondence between a P-VLAN and a VPLS instance. In
this model, the MTU-s needs to have the capability of adding the
additional P-VLAN tag to non-multiplexed ACs where customer VLANs are
not used as service delimiters. This functionality is described in
[802.1ad].
If customer VLANs need to be treated as service delimiters (e.g., the
AC is a multiplexed port), then the MTU-s needs to have the
additional capability of translating a customer VLAN (C-VLAN) to a
P-VLAN, or to push an additional P-VLAN tag, in order to resolve
overlapping VLAN tags used by different customers. Therefore, the
MTU-s in this model can be considered a typical bridge with this
additional capability. This functionality is described in [802.1ad].
The PE-rs needs to be able to perform bridging functionality over the
standard Ethernet ports toward the access network, as well as over
the PWs toward the network core. In this model, the PE-rs may need
to run STP towards the access network, in addition to split-horizon
over the MPLS core. The PE-rs needs to map a P-VLAN to a VPLS-
instance and its associated PWs, and vice versa.
The details regarding bridge operation for MTU-s and PE-rs (e.g.,
encapsulation format for Q-in-Q messages, customer's Ethernet control
protocol handling, etc.) are outside the scope of this document and
are covered in [802.1ad]. However, the relevant part is the
interaction between the bridge module and the MPLS/IP PWs in the
PE-rs, which behaves just as in a regular VPLS.
11.1. Scalability
Since each P-VLAN corresponds to a VPLS instance, the total number of
VPLS instances supported is limited to 4K. The P-VLAN serves as a
local service delimiter within the provider's network that is
stripped as it gets mapped to a PW in a VPLS instance. Therefore,
the 4K limit applies only within an Ethernet access network (Ethernet
island) and not to the entire network. The SP network consists of a
core MPLS/IP network that connects many Ethernet islands. Therefore,
the number of VPLS instances can scale accordingly with the number of
Ethernet islands (a metro region can be represented by one or more
islands).
11.2. Dual Homing and Failure Recovery
In this model, an MTU-s can be dual homed to different devices
(aggregators and/or PE-rs devices). The failure protection for
access network nodes and links can be provided through running STP in
each island. The STP of each island is independent of other islands
and do not interact with others. If an island has more than one
PE-rs, then a dedicated full-mesh of PWs is used among these PE-rs
devices for carrying the SP BPDU packets for that island. On a
per-P-VLAN basis, STP will designate a single PE-rs to be used for
carrying the traffic across the core. The loop-free protection
through the core is performed using split-horizon, and the failure
protection in the core is performed through standard IP/MPLS re-
routing.
12. Contributors
Loa Andersson, TLA
Ron Haberman, Alcatel-Lucent
Juha Heinanen, Independent
Giles Heron, Tellabs
Sunil Khandekar, Alcatel-Lucent
Luca Martini, Cisco
Pascal Menezes, Independent
Rob Nath, Alcatel-Lucent
Eric Puetz, AT&T
Vasile Radoaca, Independent
Ali Sajassi, Cisco
Yetik Serbest, AT&T
Nick Slabakov, Juniper
Andrew Smith, Consultant
Tom Soon, AT&T
Nick Tingle, Alcatel-Lucent
13. Acknowledgments
We wish to thank Joe Regan, Kireeti Kompella, Anoop Ghanwani, Joel
Halpern, Bill Hong, Rick Wilder, Jim Guichard, Steve Phillips, Norm
Finn, Matt Squire, Muneyoshi Suzuki, Waldemar Augustyn, Eric Rosen,
Yakov Rekhter, Sasha Vainshtein, and Du Wenhua for their valuable
feedback.
We would also like to thank Rajiv Papneja (ISOCORE), Winston Liu
(Ixia), and Charlie Hundall for identifying issues with the draft in
the course of the interoperability tests.
We would also like to thank Ina Minei, Bob Thomas, Eric Gray and
Dimitri Papadimitriou for their thorough technical review of the
document.
14. Security Considerations
A more comprehensive description of the security issues involved in
L2VPNs is covered in [RFC4111]. An unguarded VPLS service is
vulnerable to some security issues that pose risks to the customer
and provider networks. Most of the security issues can be avoided
through implementation of appropriate guards. A couple of them can
be prevented through existing protocols.
- Data plane aspects
- Traffic isolation between VPLS domains is guaranteed by the
use of per VPLS L2 FIB table and the use of per VPLS PWs.
- The customer traffic, which consists of Ethernet frames, is
carried unchanged over VPLS. If security is required, the
customer traffic SHOULD be encrypted and/or authenticated
before entering the service provider network.
- Preventing broadcast storms can be achieved by using routers
as CPE devices or by rate policing the amount of broadcast
traffic that customers can send.
- Control plane aspects
- LDP security (authentication) methods as described in
[RFC3036] SHOULD be applied. This would prevent
unauthenticated messages from disrupting a PE in a VPLS.
- Denial of service attacks
- Some means to limit the number of MAC addresses (per site per
VPLS) that a PE can learn SHOULD be implemented.
15. IANA Considerations
The type field in the MAC List TLV is defined as 0x404 in Section
6.2.1.
16. References
16.1. Normative References
[RFC4447] Martini, L., Rosen, E., El-Aawar, N., Smith, T., and
G. Heron, "Pseudowire Setup and Maintenance Using the
Label Distribution Protocol (LDP)", RFC 4447, April
2006.
[RFC4448] Martini, L., Rosen, E., El-Aawar, N., and G. Heron,
"Encapsulation Methods for Transport of Ethernet over
MPLS Networks", RFC 4448, April 2006.
[802.1D-ORIG] Original 802.1D - ISO/IEC 10038, ANSI/IEEE Std
802.1D-1993 "MAC Bridges".
[802.1D-REV] 802.1D - "Information technology - Telecommunications
and information exchange between systems - Local and
metropolitan area networks - Common specifications -
Part 3: Media Access Control (MAC) Bridges: Revision.
This is a revision of ISO/IEC 10038: 1993, 802.1j-1992
and 802.6k-1992. It incorporates P802.11c, P802.1p
and P802.12e." ISO/IEC 15802-3: 1998.
[802.1Q] 802.1Q - ANSI/IEEE Draft Standard P802.1Q/D11, "IEEE
Standards for Local and Metropolitan Area Networks:
Virtual Bridged Local Area Networks", July 1998.
[RFC3036] Andersson, L., Doolan, P., Feldman, N., Fredette, A.,
and B. Thomas, "LDP Specification", RFC 3036, January
2001.
[RFC4446] Martini, L., "IANA Allocations for Pseudowire Edge to
Edge Emulation (PWE3)", BCP 116, RFC 4446, April 2006.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
16.2. Informative References
[RFC4364] Rosen, E. and Y. Rekhter, "BGP/MPLS IP Virtual Private
Networks (VPNs)", RFC 4364, February 2006.
[RADIUS-DISC] Heinanen, J., Weber, G., Ed., Townsley, W., Booth, S.,
and W. Luo, "Using Radius for PE-Based VPN Discovery",
Work in Progress, October 2005.
[BGP-DISC] Ould-Brahim, H., Ed., Rosen, E., Ed., and Y. Rekhter,
Ed., "Using BGP as an Auto-Discovery Mechanism for
Network-based VPNs", Work in Progress, September 2006.
[L2FRAME] Andersson, L. and E. Rosen, "Framework for Layer 2
Virtual Private Networks (L2VPNs)", RFC 4664,
September 2006.
[L2VPN-REQ] Augustyn, W. and Y. Serbest, "Service Requirements for
Layer 2 Provider-Provisioned Virtual Private
Networks", RFC 4665, September 2006.
[RFC4111] Fang, L., "Security Framework for Provider-Provisioned
Virtual Private Networks (PPVPNs)", RFC 4111, July
2005.
[802.1ad] "IEEE standard for Provider Bridges", Work in
Progress, December 2002.
Appendix A. VPLS Signaling using the PWid FEC Element
This section is being retained because live deployments use this
version of the signaling for VPLS.
The VPLS signaling information is carried in a Label Mapping message
sent in downstream unsolicited mode, which contains the following
PWid FEC TLV.
PW, C, PW Info Length, Group ID, and Interface parameters are as
defined in [RFC4447].
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| PW TLV |C| PW Type |PW info Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Group ID |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| PWID |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Interface parameters |
~ ~
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
We use the Ethernet PW type to identify PWs that carry Ethernet
traffic for multipoint connectivity.
In a VPLS, we use a PWID (which, when using the Generalized PW ID
FEC, has been substituted with a more general identifier (AGI),
to address
extending the scope of a VPLS) to identify an emulated LAN segment.
Note that the PWID as specified in [RFC4447] is a service identifier,
identifying a service emulating a point-to-point virtual circuit. In
a VPLS, the PWID is a single service identifier, so it has global
significance across all PEs involved in the VPLS instance.
EID 4144 (Verified) is as follows:Section: Appendix A
Original Text:
In a VPLS, we use a VCID (which, when using the PWid FEC, has been
substituted with a more general identifier (AGI), to address
extending the scope of a VPLS) to identify an emulated LAN segment.
Note that the VCID as specified in [RFC4447] is a service identifier,
identifying a service emulating a point-to-point virtual circuit. In
a VPLS, the VCID is a single service identifier, so it has global
significance across all PEs involved in the VPLS instance.
Corrected Text:
In a VPLS, we use a PWID (which, when using the Generalized PW ID
FEC, has been substituted with a more general identifier (AGI),
to address
extending the scope of a VPLS) to identify an emulated LAN segment.
Note that the PWID as specified in [RFC4447] is a service identifier,
identifying a service emulating a point-to-point virtual circuit. In
a VPLS, the PWID is a single service identifier, so it has global
significance across all PEs involved in the VPLS instance.
Notes:
1. The problematic text follows a diagram depicting the PWID FEC (a.k.a. FEC-128) as it appears in RFC 4447. This diagram includes a 32-bit PWID field, but there is no VCID field. Nor is VCID mentioned anywhere in RFC 4447 - it has been used in the original Martini drafts but has then been replaced by PWID.
2. According to RFC 4447, AGI is used only in the Generalized PW ID FEC (a.k.a. FEC-129) but not in the PWID FEC (a.k.a. FEC-128).
Authors' Addresses
Marc Lasserre
Alcatel-Lucent
EMail: mlasserre@alcatel-lucent.com
Vach Kompella
Alcatel-Lucent
EMail: vach.kompella@alcatel-lucent.com
Full Copyright Statement
Copyright (C) The IETF Trust (2007).
This document is subject to the rights, licenses and restrictions
contained in BCP 78, and except as set forth therein, the authors
retain all their rights.
This document and the information contained herein are provided on an
"AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST AND
THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS
OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF
THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Intellectual Property
The IETF takes no position regarding the validity or scope of any
Intellectual Property Rights or other rights that might be claimed to
pertain to the implementation or use of the technology described in
this document or the extent to which any license under such rights
might or might not be available; nor does it represent that it has
made any independent effort to identify any such rights. Information
on the procedures with respect to rights in RFC documents can be
found in BCP 78 and BCP 79.
Copies of IPR disclosures made to the IETF Secretariat and any
assurances of licenses to be made available, or the result of an
attempt made to obtain a general license or permission for the use of
such proprietary rights by implementers or users of this
specification can be obtained from the IETF on-line IPR repository at
http://www.ietf.org/ipr.
The IETF invites any interested party to bring to its attention any
copyrights, patents or patent applications, or other proprietary
rights that may cover technology that may be required to implement
this standard. Please address the information to the IETF at
ietf-ipr@ietf.org.
Acknowledgement
Funding for the RFC Editor function is currently provided by the
Internet Society.