In essence, Spline collapses the Leaf & Spine functions – much like the Functions of Core & Aggregation Layers in a traditional 3-tier topology, would be collapsed into the Core itself in subsequent 2-Tier Topologies. The logical functions and services mapped to a particular layer are not necessarily bound to a physical device, and so are independent of the switch itself.
A key enabler of Spline architectures are the High density, multi-rate switches like the Dell EMC Networking S6100-ON.
S6100-ON’s modularity and Multi-rate capability make it a great choice for Core & Aggregation. The available Modules include 40G & 100G cards, with breakout cables enabling 50G, 25G, 10G and 1G connectivity. This means for e.g. that we can terminate everything from 25G cards (Dell EMC 14G Servers) to 1G WAN/Internet edge handover, on the same device.
When implemented as a Layer 2 VLT/MLAG topology, this will give is High Availability via Active Active LAGs & Layer 2 multi-pathing. If needed, we could peer with WAN Edge devices via Layer 3 ECMP as well.
Leaf & Spine
- Can be implemented as Layer 3 or Layer 2 (interlinks between Spine & Leaf).
- The leaf Layer advertises Edge (Server/VM) subnets into the network fabric. Spine Layer learns external/end-host subnet routes.
- Would enable both edge port count expansion (Leaf layer) as well as bandwidth/scale (Spine Layer).
- Scale out – the two layers can scale horizontally, with the total limited only by
- number of spine ports,
- over-subscription desired, &
- interconnections between them.
- If layer 3 links are used between Leaf/Spine, consider the impact on control plane and ease of management. e.g. While OSPF is widely deployed and understood, its link state overhead increases with scale.
- Merges the functions of Spine & Leaf.
- High Density Multi-rate switches terminating 1-100G.
- If the headroom and future expansion can be accommodated, then a single-tier Spline could be a viable choice.
- Interlinks could be L2 or L3. If L2, The switches could interconnect in a Multi-chassis link aggregation model (VLT) and anchor SVIs for respective VLANs at a single hop.
- Lowest Latency
- Only Two Managed Nodes