We recently replaced international MPLS with new ASA 5510s and site-to-site VPNs. However, when we deployed this we ran into a problem where each remote location has 2 ISPs for redundancy, but when enabling the VPN on both interfaces it flaps between the two and the tunnel is up and down as the tunnel gets torn down and moved between ISPs. Cisco has been working on this for 8 months now and we still don't have stable tunnels with multiple ISPs.
access-list RWS_TUNNEL remark Interesting traffic for IND-RWS tunnel access-list RWS_TUNNEL extended permit ip object-group BNG_tunnel_NETS object-group CORP_tunnel_NETS crypto map RWS_TUNNEL 1 match address RWS_TUNNEL crypto map RWS_TUNNEL 1 set peer 216.xxx.102.2 crypto map RWS_TUNNEL 1 set transform-set IND-RWS tunnel-group 216.xxx.102.2 type ipsec-l2l tunnel-group 216.xxx.102.2 ipsec-attributes pre-shared-key ***** route outside 0.0.0.0 0.0.0.0 216.xxx.206.1 1 track 2 route outside2 0.0.0.0 0.0.0.0 182.xxx.26.229 100 sla monitor 55 type echo protocol ipIcmpEcho 184.108.40.206 interface outside num-packets 5 timeout 1000 frequency 10 sla monitor schedule 55 life forever start-time now track 2 rtr 55 reachability
access-list BNG_TUNNEL remark Interesting traffic for IND-RWS tunnel access-list BNG_TUNNEL extended permit ip object-group CORP_tunnel_NETS object-group BNG_tunnel_NETS route outside2 0.0.0.0 0.0.0.0 216.xxx.102.1 crypto map BNG_TUNNEL 1 match address BNG_TUNNEL crypto map BNG_TUNNEL 1 set peer 182.xxx.26.230 216.xxx.206.4 crypto map BNG_TUNNEL 1 set transform-set L2L tunnel-group 182.xxx.26.230 type ipsec-l2l tunnel-group 182.xxx.26.230 ipsec-attributes pre-shared-key ***** tunnel-group 216.xxx.206.4 type ipsec-l2l tunnel-group 216.xxx.206.4 ipsec-attributes pre-shared-key *****
So what I've found is that when ISAKMP is enabled on both outside interfaces (remote office) and both IPs are configured as peers (central office) the VPN comes up successfully on both interfaces, but at some point will begin flapping between IPs. This is true with or without the SLA Monitoring, so even if the routes are all static, behaviour still occurs.
Any insight is appreciated.