Hi. Not sure I fully understand the question. I'll take a try. It depends on where the source and destination end points are living. Let's assume they live in different sites....we create an MP-BGP control plane connection between the spines in each site. This is just a very easy way to share MAC and IP reachability across sites. So, it the source EP is in site-1, the local spines will know the destination EP is in another site, and will use VxLAN to encapsulate the original packet and send it over to the remote spine's special TEP address (officially called the Overlay Unicast TEP) which is simply a /32 address that represents a given pod in a given site. This is routed over the Intersite Network (can also be the same as the Interpod network) which is just a generic L3 routed network. So to answer your question, in the example I give here, the packet is routed across the IPN/ISN to the remote spine and back. Hope that helps.
Excellent one
Is the routing staying within the aci multisite fabric ( routing using the spines) or is it leaving the aci fabric and then routing over?
Hi. Not sure I fully understand the question. I'll take a try. It depends on where the source and destination end points are living. Let's assume they live in different sites....we create an MP-BGP control plane connection between the spines in each site. This is just a very easy way to share MAC and IP reachability across sites. So, it the source EP is in site-1, the local spines will know the destination EP is in another site, and will use VxLAN to encapsulate the original packet and send it over to the remote spine's special TEP address (officially called the Overlay Unicast TEP) which is simply a /32 address that represents a given pod in a given site. This is routed over the Intersite Network (can also be the same as the Interpod network) which is just a generic L3 routed network. So to answer your question, in the example I give here, the packet is routed across the IPN/ISN to the remote spine and back. Hope that helps.