I have a questions regarding eiBGP Multipath. 2 iBGP speakers (R1 and R2), peering to the same PE via eBGP. Subnet behind PE is propagated to both R1 and R2, and R1 and R2 propagates that same subnet to each other. Routes are already installed in BGP.
eiBGP Multipath is enabled.
If traffic originates from, or ingresses (from inside to outside) R1, coming to subnet behind PE, how does the iBGP peer (R2) know to NOT forward the traffic back to R1?
Routers in question are ISR4431, but for lab purpose I am using CSR virtual appliances. Same IOS-XE 17.3.3. According to about 30 traceroute samples (different clients, so different flows), I have yet seen a routing loop occurs.
Hello @Timothy ACI ,
you have built a lab topology that shows the limitations of this feature eiBGP MultiPath.
By making comparable an eBGP path and an iBGP path both paths via PE and via other R node are installed.
I would say that this topology should not use eiBGP multipath to have loop free routing .
Where have you applied the command ? in a VRF or in Global routing table?
I think the feature had been introduced to be used on PE nodes and not on customer routers.
Now in IOS XE 17.3.3 it may be available also on address family ipv4 unicast but its usage needs a careful design.
Hope to help
Ibgp rtrs dont re-advertise routes learned from other ibgp rtrs due to loop prevention unless the rtr itself is a router reflector, They do however advertsie ebgp routes to each other.
Hi @Timothy ACI,
I think @Giuseppe Larosa got it right - this topology seems to cause routing loop no matter what. eiBGP should not be involved between two routers in the same AS.
The other two seemed to not get your question - this is not about routing updates, but rather how the routers handle forwarding when eiBGP multipath is enabled.
Hello @tuanquangnguyen ,
>> this is not about routing updates, but rather how the routers handle forwarding when eiBGP multipath is enabled.
This is exactly the root cause: each router R installs and treats in the same way an eBGP path to PE node and an iBGP path to other node.
Because each device performs its own CEF calculations on a given flow ( IP SA and IP DA) there are chances that for a given flow R1 sends to R2 and R2 sends back to R1 just using their own indipendent hash algoritm.
IP SA EXOR IP DA EXOR seed value and seed value is chosen at router startup.
This is why eiBGP multipath is not the right tool to use in this topology.
Hope to help