Showing results for 
Search instead for 
Did you mean: 

Spine leaf topology


I am new to the spine-leaf topology. I have a question, why do we have Border Leaf but not Border spines?


Having a Spine border will be 1 hop, having Leaf Border will be 2 hops.


I like to see spine like CORE/AGG SW in three tier art. 
and the CORE/AGG connect to SP edge router via other access SW which is "leaf" 
so finally we always end with leaf as border.



In Core/AGG tier art, SP/Internet router uses to connect the Core. Also, we cannot compare these 2 topologies. 


Core/Agg was designed for North-south traffic, whereas Spine/leaf is designed for East-west.


Hi vishal,


The Cisco ACI GOLF feature (also known as Layer 3 EVPN Services for Fabric WAN) enables much more efficient and scalable ACI fabric WAN connectivity. It uses the BGP EVPN protocol over OSPF for WAN routers that are connected to spine switches.





Guidelines and Limitations

Observe the following GOLF guidelines and limitations:

  • GOLF routers must advertise at least one route to Cisco ACI in order to accept traffic. No tunnel is created between leaf switches and the external routers until Cisco ACI receives a route from the external routers.

  • All Cisco Nexus 9000 Series ACI-mode switches and all of the Cisco Nexus 9500 platform ACI-mode switch line cards and fabric modules support GOLF. With Cisco APIC, release 3.1(x) and higher, this includes the N9K-C9364C switch.

  • At this time, only a single GOLF provider policy can be deployed on spine switch interfaces for the whole fabric.

  • Up to APIC release 2.0(2), GOLF is not supported with multipod. In release 2.0 (2) the two features are supported in the same fabric only over Cisco Nexus N9000K switches without “EX” on the end of the switch name; for example, N9K-9312TX. Since the 2.1(1) release, the two features can be deployed together over all the switches used in the multipod and EVPN topologies.

  • When configuring GOLF on a spine switch, wait for the control plane to converge before configuring GOLF on another spine switch.

  • A spine switch can be added to multiple provider GOLF outside networks (GOLF L3Outs), but the provider labels have to be different for each GOLF L3Out. Also, in this case, the OSPF Area has to be different on each of the L3extOuts and use different loopback addresses.

  • The BGP EVPN session in the matching provider L3Out in the infra tenant advertises the tenant routes defined in this L3extOut.

  • When deploying three GOLF Outs, if only 1 has a provider/consumer label for GOLF, and 0/0 export aggregation, APIC will export all routes. This is the same as existing L3extOut on leaf switches for tenants.

  • If there is direct peering between a spine switch and a data center interconnect (DCI) router, the transit routes from leaf switches to the ASR have the next hop as the PTEP of the leaf switch. In this case, define a static route on the ASR for the TEP range of that ACI pod. Also, if the DCI is dual-homed to the same pod, then the precedence (administrative distance) of the static route should be the same as the route received through the other link.

  • The default bgpPeerPfxPol policy restricts routes to 20, 000. For ACI WAN Interconnect peers, increase this as needed.

  • In a deployment scenario where there are two L3extOuts on one spine switch, and one of them has the provider label prov1 and peers with the DCI 1, the second L3extOut peers with DCI 2 with provider label prov2. If the tenant VRF has a consumer label pointing to any 1 of the provider labels (either prov1 or prov2), the tenant route will be sent out both DCI 1 and DCI 2.

  • When aggregating GOLF OpFlex VRFs, the leaking of routes cannot occur in the ACI fabric or on the GOLF device between the GOLF OpFlex VRF and any other VRF in the system. An external device (not the GOLF router) must be used for the VRF leaking.

You can find more in this link:


I hope the above helps you.



This widget could not be displayed.