One of my readers sent me an interesting question:
How does one impose a security policy on servers connected via a Clos fabric? The traditional model of segregating servers into vlans/zones and enforcing policy with a security device doesn’t fit here. Can VRF-lite be used on the mesh to accomplish segregation?
Leaf-and-spine (or Clos) architecture is just a pretty optimal way of connecting switches in a fabric that results in equidistant endpoints. Here are two slides from my Clos Fabric Explained webinar that illustrate the difference.
End-to-end bandwidth in traditional data center networks (source)
End-to-end bandwidth in a 2-tier design (source)
As long as the leaf nodes have the ability to spread the traffic across all uplinks, leaf-and-spine architecture can implement layer-2 or layer-3 networks. Layer-2 leaf-and-spine architectures may be limited to two spine nodes (in a VSS/vPC/MC-LAG cluster) unless you’re deploying one of the layer-2 ECMP technologies (Trill, FabricPath, VCS Fabric, SPB…).
You can also implement a mixed layer-2+layer-3 leaf-and-spine network with layer-3 forwarding implemented in the core switches (using any variant of multi-node FHRP) or across all leaf switches.
I would strongly recommend using layer-3 leaf-and-spine fabric with overlay virtual networks on top of it, but that’s a different story.
Connecting a security device to a leaf-and-spine fabric is no different from connecting a security device to a traditional data center network – you have to pull multiple VLANs (or VRFs) to the ToR switch to which the security device is connected.
I would virtualize the security device and connect it to overlay virtual networks, but yet again, that’s a different story.
Summary: You don’t have to learn any new tricks to implement security in leaf-and-spine fabrics. Use whatever worked for you in the past.
Need more details?
The Clos Fabrics Explained webinar contains a detailed explanation of numerous leaf-and-spine designs including:
- Non-redundant layer-3 design;
- Layer-3 design with multihomed servers;
- Layer-2-only design without server-to-switch LAG (ideal for vSphere deployments);
- Layer-2 design with server-to-switch LAG;
- Mixed L2/L3 design with L3 forwarding on spine switches;
- Mixed L2/L3 design with L3 forwarding on leaf switches.
If you need even more details, ExpertExpress might be the simplest alternative.