Spine vs Leaf Connections for Layer 3 Redundancy

[DSterling]

Current Setup:

• There is one Layer 3 (L3) link from Network B (Stacked L3 Switch) to Leaf 1A in Network A.
• This creates a single point of failure, so we want redundancy.

Proposed Plan:

• Move the existing L3 link from Leaf 1A to Spine 2.
• Add a new L3 link from Network B to Spine 1.
• After this change, Network B will have two L3 connections—one to Spine 1 and one to Spine 2 (see green lines in the diagram).

Key Question:

• Are there any concerns with connecting the Layer 3 links directly to the spine switches?
• Is it better to connect these L3 links to the spine switches (Spine 1 and Spine 2) or to additional leaf switches (e.g., Leaf 3A and Leaf 3B)?

Routing Details:

• ISIS is running on all leaf and spine switches.
• Static routes are configured on the Stacked L3 Switch pointing to Leaf 1A (and will need to be updated for the new links).
• From Network B to Network A, we can only use static routes.
• BGP is already configured between the leaf switches and the external network.

[M02@rt37 · Accepted solution]

Hello @DSterling 

Connect to the spine switches ?

Yes, as long as the spines alraedy run isis and you’re fine letting them handle external adjacencies, but this slightly breaks the “spines = transit only” design principle, no ?

It will work, but you lose the ability to apply per-tenant or security policies normaly done at the leafs, and troubleshooting become broader since spines now terminate external sessions.

It is "cleaner" to connect to additional leafs, as Border Leaf, and keep the spines pure transit.