Internet-Draft Z. Zhang
Intended status: Standards Track J. Drake
Expires: September 18, 2016 Juniper Networks, Inc.
J. Rabadan
March 17, 2016

EVPN Inter-subnet Multicast Forwarding


This document describes inter-subnet multicast forwarding procedures for Ethernet VPNs (EVPN).

Requirements Language

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 RFC2119.

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."

This Internet-Draft will expire on September 18, 2016.

Copyright Notice

Copyright (c) 2016 IETF Trust and the persons identified as the document authors. All rights reserved.

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents ( in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.

Table of Contents

1. Introduction

EVPN offers an efficient L2 VPN solution with all-active multi-homing support for intra-subnet connectivity over MPLS/IP network. EVPN also provides an integrated L2 and L3 service. When forwarding among Tenant Systems (TS) across different IP subnets is required, Integrated Routing and Bridging (IRB) can be used [ietf-bess-evpn-inter-subnet-forwarding].

An network virtualization endpoint (NVE) device supporting IRB is called a L3 Gateway. In a centralized approach, a centralized gateway provides all L3 routing functionality, and even two tenant systems on two subnets connected to the same NVE need to go through the central gateway, which is inefficient. In a distributed approach, each NVE has IRB configured, and inter-subnet traffic will be locally routed without having to go through a central gateway.

Inter-subnet multicast forwarding is more complicated and not covered in [ietf-bess-evpn-inter-subnet-forwarding]. This document describes the procedures for inter-subnet multicast forwarding.

For multicast traffic sourced from a TS in subnet 1, EVPN Broadcast, Unknow unicast, Multicast (BUM) forwarding based on RFC 7432, will deliver it to all sites in subnet 1. When IRBs in subnet 1 receive the mulitcast traffic, they deliver to other corresponding IRBs in other subnets at L3. From L3 point of view, those NVEs are routers connected to the subnet via the IRB interfaces and the source is locally attached. Nothing is different from a traditional LAN and regular IGMP/MLD/PIM procedures kick in.

If a TS is a multicast receiver, it uses IGMP/MLD to signal its interest in some multicast flows. One of the gateways is the IGMP/MLD querier for a given subnet. It sends queries out the IRB for that subnet, which in turn causes the queries to be forwarded throughout the subnet following the EVPN BUM procedures. TS's send IGMP/MLD joins via multicast, which are also forwarded throughout the subnet via EVPN BUM procedure. The gateways receive the joins via their IRB interfaces. From layer 3 point of view, again it is nothing different from a traditional LAN.

On a traditional LAN, only one router can send multicast to the LAN. That is either the PIM Designated Router (DR) or IGMP/MLD querier (when PIM is not needed - e.g., the LAN is a stub network). On the source subnet, PIM is typically needed so that traffic can be delivered to other subnets via other routers. For example, in case of PIM-SM, the DR on the source network encapsulates the initial packets for a particular flow in PIM Register messages and unicasts the Register messages to the Rendezvous Point (RP) for that flow, triggering necessary states for that flow to be built throughout the network.

That also works in the EVPN scenario, although not efficiently. Consider the example depicted in Figure 1, where a tenant has two subnets corresponding to two VLANs realized by two EVPN Instances (EVIs) at three sites. The VLAN1 and VLAN 2 shown in Figure 1 correspond to subnet 1 and subnet 2 respectively. A multicast source is located at site 1 on subnet 1 and three receivers are located at site 2 on subnet 1, site 1 and 2 on subnet 2 respectively. On subnet 1, NVE1 is the PIM DR while on subnet 2, NVE3 is the PIM DR. The connection drawn in Figure 1 among NVEs are L3 connections.

Multicast traffic from the source at site 1 on subnet 1 is forwarded to all three sites on VLAN 1 following EVPN BUM procedure. Rcvr1 gets the traffic when NVE2 sends it out of its local Attachment Circuit (AC). The three gateways for EVI1 also receive the traffic on their IRB interfaces to potentially route to other subnets. NVE3 is the DR on subnet 2 so it routes the local traffic (from L3 point of view) to subnet 2 while NVE1/2 is not the DR on subnet 2 so they don't. Once traffic gets onto subnet 2, it is forwarded back to NVE1/2 and delivered to rcvr2/3 following the EVPN BUM procedures.

Notice that both NVE1 and NVE2 receive the multicast traffic from subnet 1 on their IRB interfaces for subnet 1, but they do not route to subnet 2 where they are not the PIM DRs. Instead, they wait to receive traffic at L2 from NVE3. For example, for receiver 3 connected to NVE1 but on different IP subnet as the multicast source, the multicast traffic from source has to go from NVE1 to NVE3 and then back to NVE1 before it is being delivered to the receiver 3. This is similar to the hairpinning issue with centralized approach, here the multicast forwarding is centralized via the DR, even though distributed approach is being used for unicast (in that each NVE is supporting IRB and routing inter-subnet unicast traffic locally).

        site 1     .      site 2      .       site 3
                   .                  .
         src       .      rcvr1       .
          |        .        |         .
      --------------------------------------------  VLAN 1 (EVI1)
          |        .        |         .         |
      IRB1| DR     .    IRB1|         .     IRB1|
      IRB2|        .    IRB2|         .     IRB2| DR
          |        .        |         .         |
      --------------------------------------------  VLAN 2 (EVI2)
          |        .        |         .
         rcvr3     .       rcvr2      .
                   .                  .
        site 1     .     site 2       .      site 3

        Figure 1 - EVPN IRB multicast scenario


2. EVPN-aware Solution

This multicast hairpinning can be avoided if the following procedures are followed:

Essentially, each router on an IRB interface behaves as a DR/querier for receivers (but only the true DR behaves as a DR for sources), and multicast data traffic from IRB interfaces is limited to local receivers.

Note that link local multicast traffic (e.g. addressed to 224.0.0.x in case of IPv4)), typically use for protocols, is not subject to the above procedures and still forwarded to remote sites following EVPN procedures.

In the example in Figure 1, when NVE1 gets traffic on its IRB1 interface it will route the traffic out of its IRB2 and deliver to local rcvr3. It also sends register messages to the RP, since it is the DR on the source network. Both NVE2 and NVE3 will receive the traffic on IRB1 but neither sends register messages to the RP, since they are not the DR on the source subnet. NVE2 will route the traffic out of its IRB2 and deliver to local rcvr2. NVE3 will also route the traffic out of IRB2 even though there is no receiver at the local site, because the IGMP/MLD joins from rcvr2/3 are also received by NVE3.

2.1. IGMP/MLD Snooping Consideration

In the example in Figure 1, NVE3 receives IGMP/MLD joins from rcvr2/3 and will route packets out of IRB2, even though there are no receivers at the local site. IGMP/MLD snooping on NVE3 can prevent the traffic from actually being sent out of ACs but at L3 there will still be related states and processing/forwarding (e.g., IRB2 will be in the downstream interface list for PIM join states and forwarding routes).

To prevent NVE3 from learning those remote receivers at all, IGMP/MLD snooping on NVE3 could optionally suppress the joins from remote sites being sent to its IRB interface. With that, in the example in Figure 1, NVE3 will not learn of rcvr2/3 on IRB2 and will not try to route packets out of IRB2 at all.

2.2. Receiver sites not connected to a source subnet

In the example in Figure 1, the source subnet is connected to all NVEs that has receiver sites, and there are no receivers outside the EVPN network. As a result, PIM is not really needed and each NVE can just route multicast traffic locally. In that case, IGMP/MLD querier will be responsible to send traffic to a subnet.

If there is a receiver subnet connected to an NVE that is not connected to the source subnet, then there must exist layer 3 multicast paths between them. This could be over an L3VPN core (in this revision it is assumed that the subnets realized by EVPN are stub only and not transit) and normal PIM and MVPN procedures will be followed.

The L3VPN routes can be propagated either per RFC 4364 procedures or per EVPN Type 5 procedures [bess-evpn-prefix-advertisement]. BGP-MVPN [RFC 6514] requires that the routes used for RPF checking carry two extended communities (ECs) - VRF Route Import EC and Source AS EC. That must be applied to EVPN Prefix Advertisement (Type 5) routes as well.

2.3. Receiver sites without IRB

It is possible that a particular NVE may not have an IRB interface for its l2 domain. In that case, for traffic from another l2 domain, receivers need to receive from another NVE following EVPN procedures. The obvious choice is that it receives from the DR of that subnet. Because an NVE does not deliver traffic out of IRBs to remote sites with IRB, the DR needs to use a separate provider tunnel to deliver traffic only to sites that do not have IRB interfaces. The tunnel is advertised in new EVPN route type that is analogous to the MVPN "S-PMSI A-D" route [RFC6514]. This route will carry an EVPN Non-IRB Extended Community, indicating that a PE attached to the EVI identified in the route should join the advertised tunnel only if it does not have an IRB for that EVI. The routes could be either be a (*,*) wildcard S-PMSI A-D routes if an inclusive tunnel is used (but only for all sites without IRBs), or individual (*,g)/(s,*) S-PMSI A-D routes if selective tunnels are used per [draft-zzhang-bess-evpn-bum-procedure-updates]. The (*,*) wildcard S-PMSI A-D route may be advertised by the NVE carrying Non-IRB Site extended community for each of its BD to deliver multicast traffic routed out of the IRB interface to remote sites that do not have IRBs. Different RDs MUST be used for this (*, *) S-PMSI A-D route in the following case: if instead of an inclusive multicast Ethernet tag route, the NVE also uses (*,*) S-PMSI to deliver BUM traffic received from local ACs to remote PEs.

If [draft-sajassi-bess-evpn-igmp-mld-proxy] procedures are used, then routes from those non-IRB sites MUST also carry the EVPN non-IRB extended community, so that the DR will only forward traffic to those non-IRB NVEs.

     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
    | Type=0x06     | Sub-Type TBD  |  Flag(Octet)  | Reserved=0    |
    |                       Reserved=0                              |


The EVPN non-IRB Extended Community is a new EVPN extended community. EVPN extended communities are transitive extended community with a Type field of 6. The subtype of this new EVPN extended community will be assigned by IANA, and with the following 8-octet encoding:

The lower-order bit of the Flag is defined as non-IRB bit. A value one indicates non-IRB NVE. The rest of the undefined bits are set to zero.

2.4. Multi-homing Support

The solution works equally well in multi-homing situations, as long as the multi-homed PEs attached to the same Ethernet segment have the same IRB capability, which is expected to be the normal deployment.

As shown in Figure 2, both rcvr4 and rcvr5 are active-active multi-homed to NVE2 and NVE3. Receiver 4 is on subnet VLAN 1 and receiver 5 is on VLAN 2. When IRBs on NVE1 and NVE2 forward multicast traffic to its local attached access interface(s) based on EVPN BUM procedure, only DF for the ES deliveries multicast traffic to its multi-homed receiver. Hence no duplicated multicast traffic will be forwarded to receiver 4 or receiver 5.

         src       .        +-------- rcvr4-----+       
          |        .        |         .         |
      --------------------------------------------  VLAN 1 (EVI1)
          |        .        |         .         |
      IRB1| DR     .    IRB1|         .     IRB1|
      IRB2|        .    IRB2|         .     IRB2| DR
          |        .        |         .         |
      --------------------------------------------  VLAN 2 (EVI2)
          |        .        |         .         |
         rcvr3     .        +-------- rcvr5-----+       

        Figure 2 - EVPN IRB multicast and multi-homing

3. IANA Considerations

This document requests the following IANA assignments:

4. Security Considerations

This document does not introduce new security risks.

5. Acknowledgements

The authors would like to thank Eric Rosen for his detailed review and valuable comments.

6. References

6.1. Normative References

[I-D.sajassi-bess-evpn-igmp-mld-proxy] Sajassi, A., Patel, K., Thoria, S., Yeung, D., Drake, J. and W. Lin, "IGMP and MLD Proxy for EVPN", Internet-Draft draft-sajassi-bess-evpn-igmp-mld-proxy-00, October 2015.
[I-D.zzhang-bess-evpn-bum-procedure-updates] Zhang, J., Lin, W., Rabadan, J. and K. Patel, "Updates on EVPN BUM Procedures", Internet-Draft draft-zzhang-bess-evpn-bum-procedure-updates-01, December 2015.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997.
[RFC7432] Sajassi, A., Aggarwal, R., Bitar, N., Isaac, A., Uttaro, J., Drake, J. and W. Henderickx, "BGP MPLS-Based Ethernet VPN", RFC 7432, DOI 10.17487/RFC7432, February 2015.

6.2. Informative References

[I-D.ietf-bess-evpn-inter-subnet-forwarding] Sajassi, A., Salam, S., Thoria, S., Rekhter, Y., Drake, J., Yong, L. and L. Dunbar, "Integrated Routing and Bridging in EVPN", Internet-Draft draft-ietf-bess-evpn-inter-subnet-forwarding-00, November 2014.
[I-D.ietf-bess-evpn-prefix-advertisement] Rabadan, J., Henderickx, W., Palislamovic, S., Balus, F. and A. Isaac, "IP Prefix Advertisement in EVPN", Internet-Draft draft-ietf-bess-evpn-prefix-advertisement-01, March 2015.
[RFC4364] Rosen, E. and Y. Rekhter, "BGP/MPLS IP Virtual Private Networks (VPNs)", RFC 4364, DOI 10.17487/RFC4364, February 2006.
[RFC6514] Aggarwal, R., Rosen, E., Morin, T. and Y. Rekhter, "BGP Encodings and Procedures for Multicast in MPLS/BGP IP VPNs", RFC 6514, DOI 10.17487/RFC6514, February 2012.

Authors' Addresses

Wen Lin Juniper Networks, Inc. EMail:
Zhaohui Zhang Juniper Networks, Inc. EMail:
John Drake Juniper Networks, Inc. EMail:
Jorge Rabadan Nokia EMail: