In a typical BGP-based IP fabric (Clos architecture), leaf and spine switches are directly connected. A single point-to-point EBGP session is established over each of these links, which is a simple and scalable design. If each device in the fabric uses a unique AS number (a common design choice), a leaf switch will learn routes to a given destination from multiple spines, each with a different AS path. To enable Equal-Cost Multi-Path (ECMP) and load-balance traffic across all spines, the multipath multiple-as statement is required on the leaf switches. This command allows BGP to consider paths through different autonomous systems as equal for multipathing.

A. If all leaf devices use the same AS, BGP loop prevention will cause routes to be dropped unless as-override or allow-as in is configured.

D. EBGP is the primary routing protocol used for reachability within the IP fabric (between leaf and spine nodes), not just for external connectivity.

1. Juniper Networks TechLibrary. (2024). EVPN-VXLAN Campus Architectures. "Configure BGP in the Core and Distribution Layers". The guide states, "To enable load balancing, include the multipath multiple-as statement... This statement allows the device to load-balance traffic across multiple BGP paths that are in different autonomous systems (ASs)."

2. Juniper Networks TechLibrary. (2024). BGP User Guide. "Examples: Configuring BGP for an IP Fabric". The examples consistently show a single EBGP session between directly connected leaf and spine peers and the use of multipath multiple-as for ECMP.