1 IETF74 – L3VPN – Multicast VPN fast fail-over IETF 74 th meeting, San Francisco – L3VPN WG Multicast VPN fast fail-over draft-morin-l3vpn-mvpn-fast-failover-00 Wim Henderickx, Praveen Muley – Alcatel Lucent Thomas Morin – FT Orange Yakov Rekhter, Rahul Aggarwal – Juniper

2 IETF74 – L3VPN – Multicast VPN fast fail-over Introduction ■ Goal  Minimize connectivity restoration time for multicast VPN services in the presence of failures ■ Relying on rerouting for multicast connectivity restoration typically involves:  Waiting for unicast routing to converge  Additional signaling to restore multicast forwarding state (“join”) ■ That may not be fast enough for certain multicast applications...  Certain multicast applications often have high availability expectations ■ This work proposes tools to provide fast connectivity restoration for multicast VPN traffic:  Two complementary mechanisms providing different levels of failure coverage and restoration performance  Focus on upstream failures in the provider scope:  Upstream PE failures  Core node/link failures  (other failures too, depending on the strategy implemented)

3 IETF74 – L3VPN – Multicast VPN fast fail-over UMH selection based on tunnel status ■ UMH selection taking into account tunnel status ■ Reminder:  “UMH Selection” specifies procedures by which a downstream PE determines the PE from which it will receive a said multicast flow (C-S, C-G)  In the current spec, “UMH Selection” is done solely based on the VPN unicast routing information  Does not take into account the state of the P-tunnel that the selected UMH uses to send (C-S, C-G) to the local PE ■ Idea: let UMH procedures take into account the state of the P-tunnel from the selected/primary UMH to the PE  Make “UMH Selection” on a (downstream) PE switch to a backup (upstream) PE as soon as the (downstream) PE determines that the P-tunnel from the selected/primary (upstream) PE is down, without waiting for unicast VPN convergence  Different possible ways for a PE to detect that a P-tunnel from the selected/primary UMH to the PE is down:  P2MP OAM (Multipoint BFD)  Traffic counters  P-Tunnel signaling (RSVP-TE PathTear)  IGP tunnel root tracking  … ■ Key: avoid waiting for unicast convergence

4 IETF74 – L3VPN – Multicast VPN fast fail-over Standby BGP C-Multicast route ■ Standby BGP C-Multicast route ■ Idea : prepare the backup PE so that it is ready to become UMH when the primary PE fails ■ How ?  Besides advertising a normal (C-S,C-G) C-multicast Tree Join route to the nominal upstream PE, downstream PEs advertise a Standby C- multicast Tree Join route to the backup upstream PE  The backup upstream PE prepares for a possible failure (e.g. by joining the source)  The backup upstream PE monitors the reachability of C-S through the nominal/primary PE  On failure, traffic is forwarded by backup PE ■ Failure detection can be done, for instance:  based on P2MP OAM  based on unicast VPN reachability to C-S ■ Key: avoid additional signaling at failure time

5 IETF74 – L3VPN – Multicast VPN fast fail-over Standby C-multicast route - Example source receivers upstream CE Primary upstream PE downstream PE downstream CE Secondary upstream PE "normal" BGP C- multicast route Standby BGP C-multicast route receivers Warm standby Example ?

6 IETF74 – L3VPN – Multicast VPN fast fail-over Different protection levels ■ Cold standby  Backup PE joins towards the CE only after it detects failure of the primary PE ■ Warm standby  Backup PE is ready to send traffic when it detects the failure of the primary PE (backup PE is “pre-joined” toward the CE) ■ Hot standby  Backup PE sends the traffic before the failure occurs  Downstream PE switch to backup tunnel based on VPN unicast routes ■ Hot leaf standby  Backup PE sends the traffic before the failure occurs  Downstream PE switch to backup tunnel based on tunnel status

7 IETF74 – L3VPN – Multicast VPN fast fail-over Conclusion ■ Wrap up: two simple mechanisms to reduce connectivity restoration time for multicast traffic in a VPN context, for upstream failures in the provider scope (upstream PE & core failures) ■ UMH Selection based on P-tunnel status : avoid waiting for unicast convergence ■ Standby C-multicast route : avoid signaling at failure-time by preparing the backup upstream PE ■ These mechanisms can be used, each one alone or together, depending on the failure coverage and level of protection wanted ■ Different levels of protection: cold, warm, hot, leaf hot ■ Working group feedback ?