I’m seeking for a white paper\best practice document that can cover deploying a topology of spine-leaf data center.
The points i’m seeking clarifications are
1. underlay L3 connectivity – is IGP required to be configured between spine and leafs.
2. is multicast a must between spine and leaf for control plan operation? (forwarding BUM packets?)
3. with vxlan, are there any problems using mlags? how can i advertise a certain MAC address is available from two different VTEPs and encapsulate into vxlan from both connections?
4. in terms of configuration, how can i deploy such a scenario?
5. unconventional as it may sound, are there any issues connecting the spines directly to the network, with isis+bgp, as oppose to use some kind of edge leaf or dci leaf.
will appreciate any reference for any document\full config\explanation for those (more may arrive later)
Thanks in advanced
Please check if this document helps
1. Yes, you need some form of dynamic routing protocol between Leaf and Spine. EBGP is typically recommended for several reasons including simplicity, fast failover, “less chatty” etc., Refer to the following links for more information:
2. Multicast is not a requirement although Arista supports Multicast decapsulation to interop with third-party VTEPs. Within the Arista L3 UCN fabric, control-plane options for VxLAN are (a) HER with static flood-lists (b) CVX (c) EVPN
3. VxLAN is supported with MLAG and the MLAG’ed leaf pair seen as one logical VTEP. These links explain VxLAN Bridging and Routing operation with MLAG.
4. Above links should help.
5. You can certainly tie the Spines to other parts of your network. However, general rule of thumb is to keep the Spines “simple” with minimum number of features enabled. This promotes better stability and easier maintenance (Example: software upgrades). External Routing, DCI and services such as Firewall, Load Balancer are usually concentrated on a pair of Service Leafs.
Unable to download deployment Guide
The link does work for me; you’ll need to be logged in to access the link. Alternatively, you can go to https://www.arista.com/en/solutions/design-guides and look for “Layer 3 Leaf & Spine Design and Deployment Guide”
The benefits of BGP including fast fallover and sub-second convergence are covered in RFC7938 and L3LS Design & Deployment guide; links quoted earlier in the thread.
The key to fast convergence is running eBGP single-hop sessions over direct point-to-point Leaf-Spine routed links on 10/40/100G Ethernet which inherently offers Link Fault Signaling.
Post your Answer
You must be logged in to post an answer.