Introduction
In the data center there is more of a need for an east-west traffic architecture instead of the traditional three-layer north-south traffic architecture. This is where the spine-leaf design comes into play and a concept that works with the modern data center fabric design. Micro-services and container orchestration typically break the more traditional approach by requiring an overlay network and using VXLANs to segment across the underlay network.
Physical Topology
Traditional campus networking architecture consists of a Core, Distribution, and Access Layers. Within the spine-leaf architecture there are only two layers, the spine and the leaf. There are three rules when designing with this approach, leaf switches can never connect to leaves, spines can never connect to spines, and each leaf must connect to all of the spines in the topology.
Leaf switches act as the access layer in this topology along with the backbone being the spine switches. There are some disadvantages to the spine-leaf in which the scale is limited by the number of ports that a single spine switch has. This is a limiting factor; however, the benefits of east-west traffic patterns make up for this in more of a data center context.
Why the Spine-Leaf Architecture?
Traditional campus architecture is stopped by spanning tree which is something that comes into play by forcing 50% of the redundant links to be idle per VLAN. By having more of a spine-leaf design there is no more need for STP and layer three can be brought into the mix for better convergence times. With Rapid Per-VLAN Spanning Tree there are some more improvements to the original STP but these idle backup ports are still a drag on the networks performance.
Routing to the leaf switches unlocks the equal-cost-multi-pathing or ECMB. This is the technology that is used for edge routing cases. Having a topology that has nothing more than two hops away provides a reduction in latency and is beneficial for data center operations. There are some more types of architecture like the collapsed core where core and distribution are merged together and have an access layer.
Overlay Network
The overlay network is something that happens across the underlay network which are the physical devices and cabling. For the underlay network, there are routers and switches that the overlay network uses to transmit the virtual machines from one data center or server to another. The overlay network is something that uses VXLANs which are layer three VLANs essentially. The Virtual-Extensible Local Area Networks provide an overlay network segmentation to transfer virtual machines across the underlay network.
Layer three boundaries restrict layer two broadcasts and the way to use VXLANs then becomes available by encapsulating UDP with layer two frames. This bypasses the layer three boundary and the overlay network is usable. The control plane is then established by MP-BGP EVPN and acts as an overlay database to block flood-and-learn discovery overhead.
Conclusion
Spine-leaf architecture is needed to provide for east-west traffic in a data center for less latency where the traditional campus architecture has some bottlenecks in this approach. To make sure that you are getting up to speed with this architecture you would need to verify the underlay routing protocol. Then you would also need to audit subscription rates on the ports that a spine switch needs along with planning for an overlay mechanism like VXLANs.
This completes the discussion on the spine-leaf architecture. By providing more information about the spine-leaf architecture I encourage you to learn more with online courses, books or getting some lab time. When you get a chance try to set up a spine-leaf architecture on Packet Tracer and note where the traffic distribution works for east-west traffic. Compare this to another lab for the traditional campus architecture and note where the traffic distribution is more north-south oriented.