Copyright © 2004 Juniper Networks, Inc. Proprietary and Confidential 1 Multicast in BGP/MPLS VPNs and VPLS draft-raggarwa-l3vpn-mvpn-vpls-mcast-

Copyright © 2004 Juniper Networks, Inc. Proprietary and Confidential www.juniper.net 1

Multicast in BGP/MPLS VPNs and VPLS

draft-raggarwa-l3vpn-mvpn-vpls-mcast-01.txt

Rahul Aggarwal

[email protected]

2Copyright © 2004 Juniper Networks, Inc. Proprietary and Confidential www.juniper.net

Authors Rahul Aggarwal (Juniper) Thomas Morin (France Telecom) Luyuan Fang (AT&T) Yakov Rekhter (Juniper) Anil Lohiya (Juniper) Tom Pusateri (Juniper) Lenny Giuliano (Juniper) Chaitanya Kodeboniya (Juniper)


Agenda IP Multicast in VPLS – Issues with existing

proposals

Design Objective

Solution


IP Multicast in VPLS This talk is about IP multicast data traffic

in VPLS

It is not about VPLS control traffic

It is also not about flooding to all PEs (by the ingress PE) in the VPLS for unknown destinations for unicast traffic


Current VPLS proposals:

“Virtual Private LAN Service” (draft-ietf-l2vpn-vpls-bgp-02.txt)

“Virtual Private LAN Services over MPLS” (draft-ietf-l2vpn-vpls-ldp-05.txt )

Limitations of these solutions for IP multicast in VPLS…


VPLS Reference Model

PE 1

CE-B2

CE-B3

PE 3

PE 2

PE 4

VPLS ASite 2

CE -A1

CE-B1

CE-A2

CE-A4

CE-A3

VPLS BSite 1

VPLS BSite 2

VPLS ASite 4

Emulated Emulated LANLAN

for VPLS Bfor VPLS B

VPLS BSite 3

VSI-A

VSI-A

VSI-A

VSI-A

VSI-B

VSI-B

VSI-B

Emulated Emulated LAN for VPLSLAN for VPLS

AA

VPLS ASite 3

VPLS ASite 1

VSI – Virtual Switch Instance

No PIM peering between CEs and PEsNo PIM peering among PEs


Forwarding VPLS multicast traffic within the service provider – Emulated LAN

Ingress replication of the IP multicast packet for a given VPLS by the ingress PE

The packet is sent over the Emulated LAN associated with the VPLS

Emulated LAN is realized by ingress replication – use collection of the existing (unicast) LSPs

• From ingress PE to egress PEs

• No additional state (beyond what is require by unicast) on P routers

• May result in sending multiple copies of the same multicast packet over a given service provider link


Emulated LAN Ingress Replication: example

PE 1

CE-B2

CE-B3

PE 3

PE 2PE 4

VPLS ASite 3

CE -A1

CE-B1

CE-A2

CE-A4

CE-A3VPLS BSite 1

VPLS BSite 2

VPLS BSite 3

VSI-A

VSI-A

VSI-A

VSI-B

VSI-B

VSI-B

VPLS ASite 2

VPLS ASite 1

Links

S1G1

G1

G1

(S1, G1) traffic to Site 2(S1, G1) traffic to Site 3

The same (multicast) packet traverses link 3 times

VPLS ASite 4

VSI-A

G1

(S1, G1) traffic to Site 4


Sending multicast traffic to sites with no receivers: example

PE 1

CE-B2

CE-B3

PE 3

PE 2PE 4

VPLS ASite 3

CE -A1

CE-B1

CE-A2

CE-A4

CE-A3VPLS BSite 1

VPLS BSite 2

VPLS BSite 3

VSI-A

VSI-A

VSI-A

VSI-B

VSI-B

VSI-B

VPLS ASite 2

VPLS ASite 1

Links

S1G1

G1

G1


VPLS ASite 4

VSI-A

(S1, G1) traffic to Site 4

Multicast traffic for VPLS A extends to CE-A4,even though it has no receivers for G1

Site 4 has no receivers for G1


Sending multicast traffic to sites with no receivers

As long as PE does not keep track of IP multicast receivers within each site of a given VPLS, PE has to send IP multicast traffic to all the sites within that VPLS

As long as the ingress PE sends (multicast) traffic to all the sites within a VPLS, it is possible that the traffic will be delivered to the sites of that VPLS that have no receivers for the traffic

Suboptimal use of the service provider bandwidth due to sending IP multicast traffic to sites with no receivers is further compounded by the use of ingress replication for Emulated LAN


How to avoid sending multicast traffic to sites with no receivers – PIM/IGMP snooping

Well-known approach used by Ethernet switches

• An Ethernet switch determines whether a particular port has receivers for a given (S,G) by snooping on the PIM/IGMP messages received over that port

• Requires to disable PIM Join suppression In the context of VPLS, PE has to snoop on PIM/IGMP messages

received from:

• all sites of that VPLS (directly) connected to the PE, • Can not be avoided AND

• all the remote PEs that have members of that VPLS• Huge overhead particularly given the periodic nature of PIM

Joins

Just like with Ethernet switches, PIM/IGMP snooping in the context of VPLS requires to disable PIM Join suppression by VPLS customers


PIM snooping : example

PE 1

CE-B2

CE-B3

PE 3

PE 2PE 4

VPLS ASite 3

CE -A1

CE-B1

CE-A2

CE-A4

CE-A3VPLS B

Site 1

VPLS BSite 2

VPLS BSite 3

VSI-A

VSI-A

VSI-A

VSI-B

VSI-B

VSI-B

VPLS ASite 2

VPLS ASite 1

Links

S1G1

G1

G1


VSI-A

PE1 does not send (S1,G1) traffic to Site 4, as PE1 notices that Site 4 has no receivers for G1

PIM Join (S1, G1) from Site 2PIM Join (S1, G1) from Site 3

VPLS ASite 4 Site 4 has no receivers for G1


Multicast in 2547 VPNs vs multicast in VPLS

Focus on minimizing service provider bandwidth usage by (a) minimizing the amount of (multicast) traffic replication within the service provider, and by (b) avoiding sending traffic to the PE routers with no receivers

• At the expense of additional state within the service provider

Focus on minimizing state in the service provider routers by eliminating any multicast-related state in the P routers

• At the expense of additional bandwidth usage within the service provider

Why the tradeoffs for multicast in 2547 VPNs are NOT the same as the tradeoffs for multicast in VPLS ?

2547 VPNs: VPLS:


Solution

The existing solutions are clearly not sufficient for IP multicast support in VPLS

Draft-raggarwa-l3vpn-mvpn-vpls-mcast-00.txt• Work in progress• Aims to overcome the issues that have been

pointed out A scalable MVPN/VPLS multicast architecture

• Overcomes issues with prior solutions• This presentation talks only about VPLS

Reuse procedures across MVPN/VPLS multicast as much as possible


VPLS IP MulticastAvoid Flooding

IGMP/PIM snooping between PE and directly connected sites

• Not performed for remote sites Convert periodic PIM C-Joins snooped from a directly

connected CE to reliable protocol messages across the SP core

• Eliminates the overhead of snooping periodic PIM messages from remote sites

• This can be done using either PIM or BGP

• Draft points out the information elements The C-Join/Prune has to be sent to all the PEs in the VPLS if

the PE sending the C-Join/Prune does not know the route to the C-Source

• A database of C-Source to the PE that the C-source is behind needs to be maintained to eliminate this overhead


VPLS IP MulticastData Plane

Flexible/Scalable Transport of Customer Multicast Data Packets through the SP core

SP Multicast Trees• Allow multiple VPLSs to share a single

SP multicast tree• Can be set up using PIM or P2MP MPLS

TE LSPs or another P2MP technology Ingress Replication

• Has its applicability


Aggregate Trees Allow one SP multicast Tree to be shared across

multiple VPLSs Can be setup using PIM-SM or PIM-SSM or P2MP

MPLS TE or another P2MP technology Requires an inner label to demultiplex a particular

VPLS

• ‘Upstream’ label allocation by the root of the tree A flexible tool to reduce state in the SP network State in the SP network doesn’t grow proportional

to the number of VPLSs

• Similar to unicast in VPLS or unicast in 2547


Aggregate Trees… Use BGP signaling Mapping of an Aggregate Tree to VPLSs

associated with the Tree are signaled by the root using BGP

• Leaves of the tree are PEs belonging to all the VPLSs mapped to the tree: discovered using the auto-discovery mechanism (eg. BGP)

• Applicable to both BGP and LDP based VPLS The draft points out the information elements to

be exchanged Encoding open to discussion


Aggregate Trees: Example

PE 1

CE-B2

CE-B3

PE 3

PE 2PE 4

VPLS ASite 3

CE -A1

CE-B1

CE-A2

CE-A4

CE-A3VPLS BSite 1

VPLS BSite 2

VPLS BSite 3

VSI-A

VSI-A

VSI-A

VSI-B

VSI-B

VSI-B

VPLS ASite 2

VPLS ASite 1

Links

S1G1

G1

G1

The same (multicast) packet traverses the link only once

VPLS ASite 4

VSI-A

G1

Aggregate Tree for VPLS A and VPLS B

Upstream label for VPLS A

Upstream label for VPLS B


Aggregate Data Trees A flexible tool to create separate trees for a set of

customer groups to avoid flooding Allow one SP multicast Tree to be shared across

multiple (C-S,C-Gs) that may belong to different VPLSs

• Leaves of the tree are (C-S, C-Gs) discovered from the C-Join information

Setup using BGP signaling by the root

• Applicable to both LDP and BGP based VPLS Requires an inner label to de-multiplex a particular

VPLS

• ‘Upstream’ label allocation by the root of the tree


On P2MP MPLS TE Certainly applicable to setting up SP

multicast tree

• Potential TE benefits

Aggregate Tree and Aggregate Data Trees

Procedures in the draft are independent of the SP P2MP technology and apply to P2MP MPLS TE as well


Conclusion Solution aims to overcome the limitations

of IP multicast in VPLS with existing solutions

VPLS portion to be moved to a separate draft

WG Feedback ?

Documents

Copyright © 2004 Juniper Networks, Inc. Proprietary and Confidential 1 Multicast in BGP/MPLS VPNs and VPLS draft-raggarwa-l3vpn-mvpn-vpls-mcast-