Provider Provisioned VPN WG Hamid Ould-Brahim
Internet Draft Nortel Networks
Expiration Date: November 2004
Yakov Rekhter
Juniper Networks
(Editors)
May 2004
GVPN Services:
Generalized VPN Services using
BGP and GMPLS Toolkit
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt
Status of this Memo
This document is an Internet-Draft and is in full conformance
with all provisions of Section 10 of RFC2026 [RFC-2026], except
that the right to produce derivative works is not granted.
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 draft describes a suite of port-based Provider-provisioned
VPN services called Generalized VPNs (GVPNs) that uses BGP as a
VPN auto-discovery and GMPLS as a signaling mechanism. GVPN
services are "generalized" as the interfaces on the customers
and provider ports could be any of the interfaces supported by
Generalized MPLS (GMPLS). GVPN services outlined in this
document are: (1) a port-based Generalized Virtual Private Wire
(GVPW) where the basic unit of service is a Label Switched Path
Ould-Brahim, Rekhter November 2004 [Page 1]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
(LSP) between a pair of customers ports within a given VPN
port-topology. (2) a Generalized Virtual Private Cross-connect
(GVPXC) service where the service provider network appears to
the customer network as a GMPLS-enabled Virtual Private node. A
GVPXC service provides flexible traffic engineering on the
client network and eliminates the need for n square routing
peering between CEs. Since GVPNs uses GMPLS as the signaling
mechanism, and since GMPLS applies to both TDM and Optical
interfaces, it results that GVPN services include Optical/TDM
VPNs (though they need not be restricted to).
Original Contributors of the initial versions of this document:
Hamid Ould-Brahim (Nortel)
Yakov Rekhter (Juniper)
Luyuan Fang (AT&T)
Don Fedyk (Nortel)
Peter Ashwood-Smith (Nortel)
Eric C. Rosen (Cisco)
Eric Mannie (KPN Qwest)
John Drake (Calient Neworks)
Yong Xue (Worldcomm/UUNET)
Riad Hartani (Caspian Networks)
Dimitri Papadimitrio (Alcatel)
Lou Berger (Movaz)
1. Generalized VPN Services
Consider a service provider network that consists of devices
that supports Generalized MPLS (e.g., Optical Cross Connect,
SDH Cross Connect, etc
). We partition these devices into P
(provider) and PE (provider edge) nodes (in the context of this
document well refer to these devices as just "PE"). The P
nodes are connected only to the nodes within the providers
network (in the context of this document well refer to these
devices as just "P"). The PEs are connected to the other nodes
within the provider network (either Ps, or PEs), as well as to
the devices outside of the provider network. Well refer to
such other devices as Client Edge Devices (CEs). An example of
a CE would be a router, or an SDH cross-connect, or an Ethernet
switch.
Ould-Brahim & Rekhter. November 2004 [Page 2]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
+---+ +---+
| P | | P |
+---+ +---+
PE / \ PE
+-----+ +-----+ +--+
| | | |----| |
+--+ | | | | |CE|
|CE|----+-----+ | |----| |
+--+\ | | | +--+
\ +-----+ | |
\ | | | | +--+
\| | | |----|CE|
+-----+ +-----+ +--+
\ /
+---+ +---+
| P |....| P |
+---+ +---+
Figure 1: Generalized Port-Based VPN Reference Model
We define a "Generalized VPN" service as a Provider-provisioned
VPN service that uses BGP as a VPN auto-discovery and GMPLS as
a signaling and routing mechanisms. GVPN services are
"generalized" as the interfaces on the customers and provider
ports could be any of the interfaces supported by Generalized
MPLS (GMPLS). Since GVPN uses GMPLS as the signaling mechanism,
and since GMPLS applies to both TDM and Optical interfaces, it
results that GVPN services includes Optical/TDM VPNs (though
they need not be restricted to). Note that this draft assumes
that (1) GMPLS is used as a signaling both within the service
provider, as well as between the customer and the service
provider; (2) GMPLS is used not just as a signaling mechanism,
but as a routing mechanism within the provider network and for
services such as generalized virtual private cross-connect.
A CE is connected to a PE via one or more links. In the context
of this document a link is the same as a GMPLS Traffic
Engineering (TE) link construct, as defined in [GMPLS-ROUTING].
In the context of this document a link is a logical construct
that is used to represent grouping on a per VPN basis of
physical resources used to connect a CE to a PE. Interfaces at
the end of each link could be any of the interfaces that are
supported by GMPLS. Likewise, CEs and PEs could be any devices
that are supported by GMPLS (e.g, optical cross connects, SDH
cross-connects, LSRs, etc).
Each link may consist of one or more channels or sub-channels
(e.g., wavelength or wavelength and timeslot respectively). For
purpose of this discussion we assume that all the channels
within a given link have shared similar characteristics (e.g.,
bandwidth, encoding, etc_), and can be interchanged from the
CEs point of view. Channels on different links of a CE need not
have the same characteristics.
Ould-Brahim & Rekhter. November 2004 [Page 3]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
There may be more than one link between a given CE PE pair. A
CE may be connected to more than one PE (with at least one port
per each PE). And, of course, a PE may have more than one CE
connected to it.
If a CE is connected to a PE via multiple links and all these
links belong to the same VPN, then for the purpose of this
document these links could be treated as a single link using
the link bundling constructs [LINK-BUNDLING].
In general a link may have only data bearing channels, or only
control bearing channels, or both. For the purpose of this
discussion we assume that for a given CE-PE pair at least one
of the links between them has at least one data bearing
channel, and at least one control bearing channel, or there is
an IP connectivity between the CE and the PE that could be used
for exchanging control information (more on this in Section 4).
A link has two end-points - one on CE and one on PE. In the
context of this document we'll refer to the former as "CE
port", and to the latter as "PE port". From the above it
follows that a CE is connected to a PE via one or more ports,
where each port may consists of one or more channels or sub-
channels (e.g., wavelength or wavelength and timeslot
respectively), and all the channels within a given port have
shared similar characteristics (e.g., bandwidth, encoding,
etc_), and can be interchanged from the CEs point of view.
Channels on different ports of a CE need not have the same
characteristics. Just like links, in the context of this
document ports are logical construct that
are used to represent grouping of physical resources on a per
GVPN basis that are used to connect a CE to a PE.
At any given point in time, a given port on a PE is associated
with at most one GVPN, or to be more precise with at most one
Port Information Table (although different ports on a given PE
could be associated with different GVPNs, or to be more precise
with different Port Information Tables). This association is
established and maintained by the service provider provisioning
system.
This document assumes that the interface between the CE and PE
used for the purpose of signaling is based on GMPLS protocols
[GMPLS-RSVP-TE] and follows the procedures described in [GMPLS-
OVERLAY].
1.1 Addressing, Ports, Links, and Control Channels
This document assumes that within a given GVPN each port on a
CE that connects the CE to a PE has an identifier that is
unique within that GVPN (but need not be unique across several
GVPNs). One way to accomplish this is to assign each port an IP
address that is unique within a given GVPN, and use this
Ould-Brahim & Rekhter. November 2004 [Page 4]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
address as a port identifier. Another way to accomplish this is
to assigned each port on a CE an index that is unique within
that CE, assign each CE an IP address that is unique within a
given GVPN, and then use a tuple <port index, CE IP address> as
a port identifier.
This document assumes that within a service provider network,
each port on a PE has an identifier that is unique within that
network. One way to accomplish this would be to assign each
port on a PE an index that is unique within that PE, assign
each PE an IP address that is unique within the service
provider network (in the case of multi-provider operations, the
address has to be unique across all the providers involved),
and then use a tuple <port index, PE IP address> as a port
identifier within the provider network.
As a result, each link connecting the CE to the PE is
associated with a CE port that has a unique identifier within a
given GVPN, and with a PE port that has a unique identifier
within the service provider network. We'll refer to the former
as the customer port identifier (CPI), and to the latter as the
provider port identifier (PPI).
This document assumes that in addition to PPI, each port on PE
has also an identifier that is unique within the GVPN of that
port. One way to accomplish this is to assign each port an IP
address that is unique within a given GVPN, and use this
address as a port identifier. Another way to accomplish this is
to assign each port an index that is unique within a given PE,
assign each PE an IP address that is unique within a given GVPN
(but need not be unique within the service provider network),
and then use a tuple <port index, PE IP address> acts as a port
identifier. We'll refer to such port identifier as VPN-PPI.
Note that PE IP address used for VPN-PPI need not be the same
as PE IP address used for PPI. If for a given port on a PE its
PPI and VPN-PPI are both unnumbered, then they both could use
exactly the same port index.
Note that IP addresses used for CPIs, PPIs and VPN-PPIs could
be either IPv4 or IPv6 addresses.
For a given link connecting a CE to a PE, if CPI is an IP
address, then VPN-PPI has to be an IP address as well. And if
CPI is an <port index, CPI IP address>, then VPN-PPI has to be
an <port index, PE IP address>. However, for a given port on
PE, whether VPN-PPI of that port is an IP address or an <port
index, PE IP address> is independent of whether PPI of that
port is an IP address or an <port index, PE IP address>.
Ould-Brahim & Rekhter. November 2004 [Page 5]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
This document assumes that assignment of PPIs is controlled
solely by the service provider (without any coordination with
the GVPN customers), while assignment of CPIs and VPN-PPIs is
controlled solely by the GVPN that the CPIs and VPN-PPIs belong
to. And, of course, each GVPN could assign its CPIs and VPN-
PPIs on its own, without any coordination with other GVPNs.
This document assumes also that there is an IP control channel
between the CE and the PE. This channel could be either a
single IP hop, or an IP private network, or even an IP VPN.
Well refer to the CEs address of this channel as the CE
Control Channel Address (CE-CC-Addr), and to the PEs address
of this channel as the PE Control Channel Address (PE-CC-Addr).
Both CE-CC-Addr and PE-CC-Addr are required to be unique within
the GVPN they belong to, but are not required to be unique
across multiple GVPNs. Assignment of CE-CC-Addr and PE-CC-Addr
are controlled by the GVPN these addresses belong to.
Multiple ports on a CE could share the same control channel
only as long as all these ports belong to the same GVPN.
Likewise, multiple ports on a PE could share the same control
channel only as long as all these ports belong to the same
GVPN.
An important goal of GVPN services (particularly with respect
to GVPW and GVPXC services - see sections below) is the ability
to support what is known as "single end provisioning", where
addition of a new port to a given GVPN would involve
configuration changes only on the PE that has this port and on
the CE that is connected to the PE via this port. Another
important goal in the GVPN service is the ability to
establish/terminate an LSP between a pair of (existing) ports
within a GVPN without involving configuration changes in any of
the providers devices. The mechanisms outlined in this
document aim at achieving these goals. Specifically, as part of
the GVPN service offering, these mechanisms (1) enable the
service provider to restrict the set of ports that a given port
could be connected to, (2) enable the service provider to
provide a CE with the information about the ports that the CE
could be connected, (3) enable a CE to establish the actual LSP
to a subset of ports provided by (2). Finally, the mechanisms
allow different GVPN topologies to be supported ranging from
hub-and-spoke to complete mesh.
2. Port-based Generalized Virtual Private Wire (GVPW)
A Generalized Virtual Private Wire (GVPW) is a port-based
VPN service where a pair of CEs could be connected through
the service provider network via a GMPLS-based LSP within a
given VPN port topology. It is precisely this LSP that forms
the basic unit of the GVPW service that the service provider
network offers. If a port by which a CE is connected to a PE
consists of multiple channels (e.g., multiple wavelengths), the
Ould-Brahim & Rekhter. November 2004 [Page 6]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
CE could establish LSPs to multiple other CEs over this single
port.
The service provider does not initiate the creation of an
LSP between a pair of PE ports. This is done rather by the
CEs, which attach to the ports. However, the SP, by using
the mechanisms/toolkit outlined in this document, restricts
the set of other PE ports, which may be the remote endpoints
of LSPs that have the given port as the local endpoint.
Subject to these restrictions, the CE-to-CE connectivity is
under the control of the CEs themselves. In other words, SP
allows a GVPN to have a certain set of topologies (expressed
as a port-to-port connectivity matrix), and CE-initiated
signaling is used to choose a particular topology from that
set.
A PE maintains for each GVPW configured on that PE a port
information tables (PIT) associated with each GVPW that has at
least one port configured on a PE. A PIT contains a list of
<CPI, PPI> tuples for all the ports within its GVPN. Note that
a PIT may as well hold routing information (for example when
CPIs are learnt using a routing protocol).
PE PE
+---------+ +--------------+
+--------+ | +------+| | +----------+ | +--------+
| VPN-A | | |VPN-A || | | VPN-A | | | VPN-A |
| CE1 |--| |PIT || BGP route | | PIT | |-| CE2 |
+--------+ | | ||<----------->| | | | +--------+
| +------+| Distribution| +----------+ |
| | | |
+--------+ | +------+| | +----------+ | +--------+
| VPN-B | | |VPN-B || -------- | | VPN-B | | | VPN-B |
| CE1 |--| |PIT ||-( GMPLS )--| | PIT | |-| CE2 |
+--------+ | | || (Backbone ) | | | | +--------+
| +------+| --------- | +----------+ |
| | | |
+--------+ | +-----+ | | +----------+ | +--------+
| VPN-C | | |VPN-C| | | | VPN-C | | | VPN-C |
| CE1 |--| |PIT | | | | PIT | |-| CE2 |
+--------+ | | | | | | | | +--------+
| +-----+ | | +----------+ |
+---------+ +--------------+
Figure 2 Generalized Virtual Private Wire
2.1 VPN Auto-discovery Mechanism
This document assumes a BGP-based auto-discovery for supporting
GVPW services.
Ould-Brahim & Rekhter. November 2004 [Page 7]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
A PIT on a given PE is populated from two sources: the
information related to the CEs ports attached to the ports on
that PE (this information could be optionally received from the
CEs), and the information received from other PEs. Well refer
to the former as the "local" information, and to the latter as
the "remote" information.
Propagation of local information to other PEs is accomplished
by using BGP VPN auto-discovery procedures, as specified in
[BGP-VPN-AUTODISCOVERY]. To restrict the flow of this
information to only the PITs within a given GVPN, we use BGP
route filtering based on the Route Target Extended Community
[BGP-COMM], as follows.
Each PIT on a PE is configured with one or more Route Target
Communities, called "export Route Targets", that are used for
tagging the local information when it is exported into
providers BGP. The granularity of such tagging could be as
fine as a single <CPI, PPI> pair. In addition, each PIT on a PE
is configured with one or more Route Target Communities, called
"import Route Targets", that restrict the set of routes that
could be imported from providers BGP into the PIT to only the
routes that have at least of these Communities.
When a service provider adds a new GVPN port to a particular
PE, this port is associated at provisioning time with a PIT on
that PE, and this PIT is associated (again at provisioning
time) with that GVPN.
Once a port is configured on the PE, the CE that is attached
via this port to the PE MAY pass to the PE the CPI information
of that port. This document assumes that this is accomplished
by using BGP (however, the document doesnt preclude the use
of other mechanisms).
This information, combined with the PPI information available
to the PE, enables the PE to create a tuple <CPI, PPI> for such
port, and then use this tuple to populate the PIT of the GVPN
associated with that port.
In order to establish an LSP, a CE needs to identify all other
CEs in the CE's GVPN it wants to connect to. A CE may already
have obtained the CE list through configuration or through some
other schemes (such schemes are outside the scope of this
draft).
Ould-Brahim & Rekhter. November 2004 [Page 8]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
A port, in addition to its CPI and PPI may also have other
information associated with it that describes characteristics
of the channels within that port, such as encoding supported by
the channels, bandwidth of a channel, total unreserved
bandwidth within the port, etc. This information could be
further augmented with the information about certain
capabilities of the Service Provider network (e.g., support
RSOH DCC transparency, arbitrary concatenation, etc
). This
information is used to ensure that ports at each end of an LSP
have compatible characteristics, and that there are sufficient
unallocated resources to establish an LSP. Distribution of this
information (including the mechanisms for distributing this
information) is identical to the distribution of the <CPI, PPI>
information. Distributing changes to this information due to
establishing/terminating of LSPs is identical to the
distribution of the <CPI, PPI> information, except that
thresholds should be used to contain the volume of control
traffic caused by such distribution.
It may happen that for a given pair of ports within a GVPN,
each of the CEs connected to these ports would concurrently try
to establish an LSP to the other CE. If having a pair of LSPs
between a pair of ports is viewed as undesirable, the way to
resolve this is to require the CE with the lower value of CPI
to terminate the LSP originated by the CE. This option could be
controlled by configuration on the CE devices.
2.1.1 Encoding of CPI, PPI, and channel characteristics in BGP
The <CPI, PPI> mapping is carried using the Multiprotocol
Extensions BGP [RFC2858]. [RFC2858] defines the format of two
BGP attributes, MP_REACH_NLRI and MP_UNREACH_NLRI that can be
used to announce and withdraw the announcement of reachability
information. We introduce a new address family identifier (AFI)
for GVPN (to be assigned by the IANA), a new subsequent address
family identifier (to be assigned by the IANA), and also a new
NLRI format for carrying the CPI and PPI information.
One or more <PPI, CPI> tuples could be carried in the above
mentioned BGP attributes.
The format of encoding a single <PPI, CPI> tuple is shown in
Figure 3 below:
Ould-Brahim & Rekhter. November 2004 [Page 9]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
+---------------------------------------+
| Length (1 octet) |
+---------------------------------------+
| PPI AFI (2 octets) |
+---------------------------------------+
| PPI Length (1 octet) |
+---------------------------------------+
| PPI (variable) |
+---------------------------------------+
| CPI AFI (2 octets) |
+---------------------------------------+
| CPI (length) |
+---------------------------------------+
| CPI (variable) |
+---------------------------------------+
Figure 3: NLRI BGP encoding
The use and meaning of these fields are as follows:
Length:
A one octet field whose value indicates the length of
the <PPI, CPI> Information tuple in octets.
PPI AFI:
A two octets field whose value indicates address
family identifier of PPI
PPI Length:
A one octet field whose value indicates the length of
of the PPI field
PPI field:
A variable length field that contains the value of
the PPI (either an address or <port index,
address> tuple
CPI AFI field:
A two octets field whose value indicates address
family of the CPI.
CPI Length:
A once octet field whose value indicates the
length of the CPI field.
CPI (variable):
A variable length field that contains the CPI
Ould-Brahim & Rekhter. November 2004 [Page 10]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
value (either an address or <port index, address> tuple.
2.2 Signaling
Once a CE obtains the information about the CPIs of other ports
within the same GVPN, which we'll refer to as "target ports",
the CE uses a (subset of) GMPLS signaling, to request the
provider network to establish an LSP to a target port.
For inter-CE connectivity, the request originated by the CE
contains the CPI of the port on the CE that CE wants to use for
the LSP, and the CPI of the target port. When the PE attached
to the CE that originated the request receives the request, the
PE identifies the appropriate PIT, and then uses the
information in that PIT to find out the PPI associated with the
CPI of the target port carried in the request. The PPI should
be sufficient for the PE to establish an LSP. Ultimately the
request reaches the CE associated with the target CPI (note
that the request still carries the CPI of the CE that
originated the request). If the CE associated with the target
CPI accepts the request, the LSP is established.
Note that a CE need not establish an LSP to every target port
that CE knows about - it is a local to the CE matter to select
a subset of target ports to which the CE will try to establish
LSPs.
When a CE sends an RSVP Path message to a PE, the source IP
address in the IP packet that carries the message is set to the
appropriate CE-CC-Addr, and the destination IP address in the
packet is set to the appropriate PE-CC-Addr. When the PE sends
back to the CE the corresponding Resv message, the source IP
address in the IP packet that carries the message is set to the
PE-CC-Addr, and the destination IP address is set to the CE-CC-
Addr.
Likewise, when a PE sends an RSVP Path message to a CE, the
source IP address in the IP packet that carries the message is
set to the appropriate PE-CC-Addr, and the destination IP
address in the packet is set to the appropriate CE-CC-Addr.
When the CE sends back to the PE the corresponding Resv
message, the source IP address in the IP packet that carries
the message is set to the CE-CC-Addr, and the destination IP
address is set to the PE-CC-Addr.
In addition to being used for IP addresses in the IP packet
that carries RSVP messages between CE and PE, CE-CC-Addr and
PE-CC-Addr are also used in the Next/Previous Hop Address field
of the IF_ID RSVP_HOP object that is carried between CEs and
PEs.
In the case where a link between CE and PE is a numbered non-
bundled link, the CPI and VPN-PPI of that link are used for the
Type 1 or 2 TLVs of the IF_ID RSVP HOP object that is carried
Ould-Brahim & Rekhter. November 2004 [Page 11]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
between the CE and PE. In the case where a link between CE and
PE is an unnumbered non-bundled link, the CPI and VPN-PPI of
that link are used for the IP Address field of the Type 3 TLV.
In the case where a link between CE and PE is a bundled link,
the CPI and VPN-PPI of that link are used for the IP Address
field of the Type 3 TLVs.
When a CE originates a Path message to establish an LSP from a
particular port on that CE to a particular target port the CE
uses the CPI of its port in the Sender Template object. If the
CPI of the target port is an IP address, then the CE uses it in
the Session object. And if the CPI of the target port is a
<port index, IP address> tuple, then the CE uses the IP address
part of the tuple in the Session object, and the whole tuple as
the Unnumbered Interface ID subobject in the ERO. When the Path
message arrives at the ingress PE, the PE selects the PIT
associated with the GVPN, and then uses this PIT to map CPIs
carried in the Session and the Sender Template objects to the
appropriate PPIs. Once the mapping is done, the ingress PE
replaces CPIs with these PPIs. As a result, the Session and the
Sender Template objects that are carried in the GMPLS signaling
within the service provider network carry PPIs, and not CPIs.
At the egress PE, the PE performs the reverse mapping it maps
PPIs carried in the Session and the Sender Template object into
the appropriate CPIs, and then sends the Path message to the CE
that has the target port.
2.3 GVPW Routing Considerations
It is also desirable, that the service provider, as a value
added service, may provide to a GVPW-based CE with a list of
ports on all other CEs that belong to the same VPN. This is
accomplished by passing the information stored in the PE PITs
to the attached CE. A way to accomplish this is by using BGP
Multi-protocol extensions (however this draft doesn't preclude
other mechanisms to be used). Although optional, this draft
recommends the PE to signal to the attached CEs the remote CPIs
it learnt from the remote CEs part of the same GVPN. A CE may
decide to initiate an LSP setup request to a remote CE only
when it learns the CPI of the remote CE from the PE. This has
the benefit to avoid rejecting LSP setup request while the PE
is populating the PITs.
3. Generalized Virtual Private Cross-Connect (GVPXC)
A GVPXC is a GVPN service where the service provider network
appears as a virtual private cross-connect. A GVPXC operates
similarly to a physical optical cross-connect except that it
applies to GMPLS-based interfaces and allows a wide spectrum of
port topology such as hub and spoke, full mesh, and arbitrary
topologies. The GVPXC port topology is defined by the customer,
and enforced by the service provider. Customers can signal any
inter-port connectivity according to the topology implemented by
the VPOXC. Client devices operate within the VPOXC space
independently from the service provider network operations.
Ould-Brahim & Rekhter. November 2004 [Page 12]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
GVPXC
+-------------------------------+
| +---+ +---+ |
| | P |....| P | |
| +---+ +---+ |
| PE / \ PE |
| +-----+ +-----+ | +--+
| | | | |-|--| |
+--+ | | | | | | |CE|
|CE|--|-+-----+ | |-|--| |
+--+\ | | | | | +--+
\| +-----+ | | |
| | | | | | +--+
|\| | | |-|--|CE|
| +-----+ +-----+ | +--+
| \ / |
| +---+ +---+ |
| | P |....| P | |
| +---+ +---+ |
| |
+-------------------------------+
Figure 4: GVPXC Reference Model
The bandwidth associated with each GVPXC depends on the access
bandwidth of each CE to the GVPXC and the port topology
implemented within the GVPXC. As sites are added or removed to
the GVPXC, the total GVPXC bandwidth is accordingly adjusted.
The basic unit of the GVPXC service is a GMPLS LSP between a
port on one CE and a port on another CE crossing the GVPXC
node. In the case of TDM LSP, rules are driven by [GMPLS-SONET-
SDH] for SDH/Sonet interfaces. These rules must be used when
establishing TDM connections from CE-port(s) to CE-port(s) over
the GVPXC. The number of ports depends on the concatenation
capabilities of these interfaces keeping in mind that when
provided, virtual concatenation does not constraint the GVPXC
port capability. If a port on CE has multiplexing capabilities,
the same port could be used to connect to more than one
(remote) CE ports.
A GVPXC port can be moved to another PE port (or even to
another PE) without changing the GVPXC addressing used by the
customer to request connectivity. Addition/Deletion/Changes of
the VPOXC port addresses requires no coordination with the
service provider addressing scheme. GVPXC may be used by a
customer to exchange customers GMPLS routing information
related to the customers network, as from customers point of
view (and specifically from customers routing/signaling point
of view) the service appears as a single GMPLS-capable node.
Ould-Brahim & Rekhter. November 2004 [Page 13]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
3.1 GVPXC Routing Considerations
From a customers point of view a GVPXC can be deployed in one
of the two deployment scenarios:
a) with off-line path computation or
b) with on-line path computation
In off-line path computation mode, an off-line tool is used by
the customer to compute paths for all LSPs that cross the GVPXC
node. Each node within the private network is provided with the
outcome of computation for the LSP that cross the GVPXC and are
originated by the node.
On-line path computation assumes that the GVPXC node
participates in the GMPLS routing with customers network , or
to be more precise, participates in flooding GMPLS routing
information of the client to whom that node belongs.
GVPXC-A
+-----------------------------------------+
| PE1 PE2 |
| +-----------+ +-----------+ |
+-----+ VPN-LSP| | | | | +-----+
|CE1-A|<--------->+------+ GVSI-LSP | +------+ | | |CE2-A|
+-----+ | | |GVSI-A| |<---------->| |GVSI-A|<---->+-----+
| | +------+ | | +------+ | |
+-----------------------------------------+
| | | |
| | GVPXC-B | |
+-----------------------------------------+
+-----+ VPN-LSP| +------+ | | +------+ | | +-----+
|CE1-B|<--------->|GVSI-B| | GVSI-LSP | |GVSI-B|<---->|CE2-B|
+-----+ | | +------+ |<---------->| +------+ | | +-----+
| | | | | |
| +-----------+ +-----------+ |
| |
+-----------------------------------------+
Figure 5: Anatomy of the GVPXC
In order for the GVPXC to participate in GMPLS routing with the
customers network, the GVPXC needs to a) establish a routing
adjacency with attached CEs, b) generate routing information
with traffic engineering (TE) information for the set of CE-PE
TE-links attached to the GVPXC, and c) floods TE-Link routing
information (such as the ones learnt from other customers
network nodes) to the attached CEs using normal GMPLS routing
procedures.
Ould-Brahim & Rekhter. November 2004 [Page 14]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
To accomplish the above steps, each PE maintains for each GVPXC
service VPN information tables. We refer to such information as
Generalized Virtual Switching Instance (GVSI). A GVSI can be
viewed as a combination of GVPXC Routing and Forwarding tables
and GVPXC Port information Table. GVSIs associated with one
GVPXC are inter-connected by tunnel-based control channels. One
realization of the control channel between a pair of GVSI is to
use an IP/MPLS-based tunnels where plain private IGP adjacency
can be established. Note that such adjacency is only used for
distributing customer's routing information among the GVSIs.
When receiving routing updates from the CE neighbors, the PE
(or more precisely the GVSI configured on that PE) updates its
IGP database and propagates the updates to other GVSIs using
basic IGP procedures across the tunnel-based control-channels.
The approach for distributing private reachability is similar
to the virtual router approach used in layer-3 VPNs with the
exception that a) the tunnel-based control channels are not
visible to the CE and b) since the GVPXC represents a virtual
node, the GVSIs will advertise VPN routing updates with the
same GVPXC ROUTER_ID.
3.2 Auto-Discovery
VPN auto-discovery procedures described in [BGP-VPN-AUTO-
DISCOVERY] are used to enable the PEs to determine which GVSIs
are in the same GVPXC. Once the GVSIs are reachable through the
control-based tunnels, private routes are then exchanged by
running an instance of routing protocol per pair of GVSIs
basis.
Carrying GVSIs information in BGP-MP is done as follows. The
NLRI address prefix is an address of one of the GVSIs
configured on the PE.
BGP Route target extended community is used to constrain route
distribution between PEs (GVSIs). The BGP Next hop carries the
service provider control-channel tunnel endpoint address which
is in the service provider addressing space.
In addition to GVSI related information, NLRI will also carry
the tuples <CPIs, PPIs> as described in section 2.1.1.1. This
information is used to establish end to end LSP between CEs
across the GVPXC node (see section below).
3.3 Signaling
An LSP initiated within the VPN domain may contain a path that
crosses the GVPXC node. We refer to the LSP that crosses the
GVPXC node as a VPN-LSP. The creation/termination of a VPN-LSP
could be driven either by mechanisms outside of GMPLS (e.g.,
via configuration control on the CE), or by mechanisms within
GMPLS (e.g., as a result of the CE at the head-end of the VPN-
LSP receiving LSP setup requests originated by some other LSRs
within the VPN space).
Ould-Brahim & Rekhter. November 2004 [Page 15]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
A CE may decide to use the VPN-LSP as a forwarding Adjacency
(FA) using procedures described in [LSP-HIERARCHY], and
announces this LSP as a Traffic Engineering (TE) link into the
same instance of the CE GMPLS control plane (or more precisely
CE ISIS/OSPF component) as the one that was used to create the
VPN-LSP. In this case, ISIS/OSPF floods the information about
VPN-LSP just as it floods the information about any other
links. As a result of this flooding, an LSR within the VPN has
in its TE link state database the information about not just
basic TE links (from other nodes including GVPXC TE-links), but
VPN-LSPs as well.
In order to establish the VPN-LSPs, the GVSIs/PEs are inter-
connected at the data-plane level through GMPLS-based LSPs. We
refer to such LSPs as GVSI-LSPs (see figure 5). A GVSI-LSP is
either pre-configured or constructed dynamically as a result of
a PE receiving a VPN-LSP PATH message. A given GVSI-LSP may
map exactly to one VPN-LSP or to many VPN-LSPs. When a GVSI-LSP
is created dynamically, its attributes are inherited from the
VPN-LSP, which induced its creation and from the information
maintained in the port information table associated with the
GVSI. And for provisioned GVSI-LSPs, a policy-based mechanism
may be needed to associate attributes to the GVSI-LSPs.
Note that the bandwidth of the GVSI-LSP must be at least as big
as the LSP that induced it, but may be bigger if only discrete
bandwidths are available for the GVSI-LSP.
Upon receiving the VPN-LSP PATH message, the ingress PE must
then determine the egress PE using the GVSI IGP database and
the PIT table or just the PIT table (in case the ERO contains
already the destination CPI corresponding to an existing entry
in the PIT table)The PE then tries to find an existing GVSI-LSP
between the ingress PE and the egress PE .
If a match is found, where the GVSI-LSP has enough unreserved
bandwidth for the VPN-LSP being signaled, and the G-PID of the
GVSI-LSP is compatible with the G-PID of the VPN-LSP being
signaled, the PE uses that GVSI-LSP.
Otherwise (if no existing GVSI-LSP is found), the PE sets up a
new GVSI-LSP. That is, it initiates a new LSP setup just for
the GVSI-LSP. Once the GVSI-LSP is established, the PE
encapsulates the original VPN-LSP PATH message in an IP tunnel,
and unicasts the message to the tail end of the GVSI-LSP.
The Path message for the original VPN-LSP MUST contain an IF_ID
RSVP_HOP object instead of an RSVP_HOP object; and the data
interface identification MUST identify the GVSI-LSP. The
ingress PE adjusts the ERO of the VPN-LSP path message and
sends it to the egress PE of the GVSI-LSP, not to the next hop
along the GVSI-LSP's path.
Ould-Brahim & Rekhter. November 2004 [Page 16]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
The egress PE will process the VPN-LSP using normal GMPLS
signaling procedures and sends it to the egress CE. VPN-LSPs
are then nested across the GVSI-LSPs.
4. Others Issues
o One vs more than one GVPN
The solution described in this document requires each customer
port to be in at most one GVPN, or to be more precise requires
each customer port connected to a given PE to be associated
with at most one PIT on that PE. It has been asserted that this
requirement is too restrictive, as it doesnt allow to realize
certain connectivity scenarios. To understand why this
assertion is incorrect wed like to make several observations.
First, the solution/mechanisms described in this document
allows control connectivity between customers ports at the
granularity of individual ports. This is because each local
port on a PE could have its own PIT (GVSI), and the granularity
of the information that is used to populate this PIT could be
as fine as a single remote port (port on some other PE).
Second, ports that are present in a given PIT need not have the
same administrative control. For example, some ports in a given
PIT may belong to the same organization (have the same
administrative control) as the local ports associated with that
PIT, while some other ports in exactly the same PIT may belong
to organizations different from the one associated with the
local ports. In that sense, a single PIT could combine both an
Intranet and an Extranet.
As a result, it should be abundantly obvious to the informed
reader that the solution described in this document allows to
realize any arbitrary inter-port connectivity matrix.
Therefore, no other solution could be less restrictive than
then one described in this document.
o Exchanging VPN-ID between CE and PE
The solution described in this document assumes that an
association of a particular port on a CE with a particular GVPN
(or to be more precise with a particular PIT on a PE) is done
by the GVPN service provider, as part of the provisioning the
port on the PE (associating the PEs port with a particular
PIT, and connecting the CEs port with the PEs port). Once
this association is established, the CE could request
establishment of an LSP to any customers port present in the
PIT. Important to note that in order to select a particular
Ould-Brahim & Rekhter. November 2004 [Page 17]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
port within the PIT for the purpose of establishing an LSP to
that port the only information that the CE needs to identify
that port is the CPI of that port. Also important to note that
the CPI is either an IP address, or a combination of
<portindex, IP address>, but it doesnt include any such thing
as VPN-ID.
Therefore, the solution described in this document doesnt
involve exchanging VPN-IDs between CE and PE in (GMPLS)
signaling. Moreover, the lack of exchanging VPN-ID in signaling
has no adverse effect on the ability to support any arbitrary
inter-port connectivity matrix, and more generally on the
flexibility of the solution described in this draft.
o Multiple Routing Domains
Since the protocol used to populate a PIT with remote
information is BGP, since BGP works across multiple routing
domains, and since GMPLS signaling isnt restricted to a single
routing domain, it follows that the mechanisms described in
this document could support an environment that consists of
multiple routing domains.
o Addressing
The mechanisms described in this document allow for a wide
range of choices with respect to addresses used for CPI, PPI,
and VPN-PPI. For example, one could use either IPv4 addresses,
or IPv6 addresses, or NSAPs. Different GVPN customers of a
given service provider may use different types of addresses.
Moreover, different GVPNs attaching to the same PE may use
different addressing schemes. The types of addresses used for
PPIs within a given service provider network are independent
from the type of addresses used for CPI and VPN-PPI by the GVPN
customers of that provider.
o GVPNs and Layer-2/3 VPNs
While in the context of this document a CE is a device that
uses the GVPN service, such a device, in turn, could be used to
offer VPN services (e.g., RFC2547, Virtual Routers, Layer 2
VPNs) to other devices (thus becoming a PE with respect to
these devices). Moreover, a CE device that uses the GVPN
service could, in turn be used to offer GVPN services to other
devices (thus becoming a PE with respect to these devices).
5. Security Considerations
Since association of a particular port with a particular GVPN
(or to be more precise with a particular PIT) is done by the
service provider as part of the service provisioning process
(and thus can't be altered via signaling between CE and PE),
and since signaling between CE and PE is assumed to be over a
Ould-Brahim & Rekhter. November 2004 [Page 18]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
private network (and thus can't be spoofed by entities outside
the private network), the solution described in this document
doesn't require authentication in signaling.
6. References
[BGP-VPN-AUTODISCOVERY] Ould-Brahim, H., Rosen, E., Rekhter,
Y., "Using BGP as an Auto-Discovery Mechanism for Network-
based VPNs", work in progress
[GMPLS-SIGNALING] Berger, L. (editor), "Generalized MPLS -
Signaling Functional Description", January 2003, RFC3471.
[GMPLS-RSVP-TE] Berger, L. (editor), "Generalized MPLS
Signaling - RSVP-TE Extensions", RFC3473, January 2003.
[GMPLS-ROUTING] Kompella, K., Rekhter, Y., "Routing Extensions
in Support of Generalized MPLS", work in progress
[GMPLS-HIERARCHY] Kompella, K., Rekhter, Y., "LSP Hierarchy
with Generalized MPLS TE", work in progress.
[LINK-BUNDLING] Kompella, K., Rekhter, Y., Berger, L., "Link
Bundling in MPLS Traffic Engineering", work in progress.
[GVPN-REQ] Ould-Brahim, H., Rekhter, Y., et al., "Service
Requirements for Optical Virtual Private Networks", work in
progress, July 2001.
[GMPLS-OVERLAY] Swallow, G., et al., "GMPLS RSVP Support for
the Overlay Model", work in progress.
7. Author's Addresses
Hamid Ould-Brahim
Nortel Networks
P O Box 3511 Station C
Ottawa ON K1Y 4H7 Canada
Phone: +1 (613) 765 3418
Email: hbrahim@nortelnetworks.com
Yakov Rekhter
Juniper Networks
1194 N. Mathilda Avenue
Sunnyvale, CA 94089
Email: yakov@juniper.net
Ould-Brahim & Rekhter. November 2004 [Page 19]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
Don Fedyk
Nortel Networks
600 Technology Park
Billerica, Massachusetts
01821 U.S.A
Phone: +1 (978) 288 3041
Email: dwfedyk2nortelnetworks.com
Peter Ashwood-Smith
Nortel Networks
P.O. Box 3511 Station C,
Ottawa, ON K1Y 4H7, Canada
Phone: +1 613 763 4534
Email: petera@nortelnetworks.com
Eric C. Rosen
Cisco Systems, Inc.
250 Apollo drive
Chelmsford, MA, 01824
E-mail: erosen@cisco.com
Eric Mannie
KPNQwest
Terhulpsesteenweg 6A
1560 Hoeilaart
Belgium
Phone: +32 2 658 56 52
Email: eric.mannie@ebone.com
Luyuan Fang
AT&T
200 Laurel Avenue
Middletown, NJ 07748
Email: Luyuanfang@att.com
Phone: +1 (732) 420 1920
John Drake
Calient Networks
5853 Rue Ferrari
San Jose, CA 95138
USA
Phone: +1 408 972 3720
Email: jdrake@calient.net
Yong Xue
UUNET/WorldCom
Ashburn, Virginia
(703)-886-5358
yxue@uu.net
Riad Hartani
Caspian Networks
Ould-Brahim & Rekhter. November 2004 [Page 20]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
170 Baytech Drive
San Jose, CA 95143
Phone: 408 382 5216
Email: riad@caspiannetworks.com
Dimitri Papadimitrio
Alcatel
Francis Wellesplein 1,
B-2018 Antwerpen, Belgium
Phone: +32 3 240-8491
Email: Dimitri.Papadimitriou@alcatel.be
Lou Berger
Movaz Networks, Inc.
7626 jones Branch Drive, Suite 615
McLean, VA 22102
Phone: +1 703 847 1801
Email: lberger@movaz.com
Ould-Brahim & Rekhter. November 2004 [Page 21]
draft-ouldbrahim-ppvpn-gvpn-bgpgmpls-05.txt May 2004
Full Copyright Statement
Copyright (C) The Internet Society (2000). All Rights Reserved.
This document and translations of it may be copied and
furnished to others, and derivative works that comment on or
otherwise explain it or assist in its implementation may be
prepared, copied, published and distributed, in whole or in
part, without restriction of any kind, provided that the above
copyright notice and this paragraph are included on all such
copies and derivative works. However, this document itself may
not be modified in any way, such as by removing the copyright
notice or references to the Internet Society or other Internet
organizations, except as needed for the purpose of developing
Internet standards in which case the procedures for copyrights
defined in the Internet Standards process must be followed, or
as required to translate it into languages other than English.
The limited permissions granted above are perpetual and will
not be revoked by the Internet Society or its successors or
assigns.
Ould-Brahim & Rekhter. November 2004 [Page 22]