Spine and Leaf Topology

---

Spine and leaf topology is a network topology commonly used in datacenters comprising a top tier of aggregation switches forming a backbone for a leaf tier of top-of-rack switches.

• better support for
  • east-west traffic
  • SDN and overlay network within datacenters
• two layers:
  • spine layer
    • comprises a backbone of top-tier distribution switches
    • not linked to one another
  • leaf layer
    • contains access switches
    • Each access switch is connected to every spine switch in a full mesh topology
    • access switches never have direct connections to one another

Benefits

• each server is only ever a single hop from the backbone
  • makes network latency more predictable
• multiple redundant paths between a leaf switch and the backbone
  • allows for load balancing and failover
• no direct connections between spine switches in the backbone or between leaf switches
  • network is loop free and does not need to run spanning tree
  • Instead, each leaf switch runs a protocol called Equal Cost Multipathing (ECMP)
    • distribute traffic between the links to the spine switches
• Servers are connected to multiple leaf switches for multipath redundancy, using a first hop gateway protocol to determine the active path
• improved scalability
  • adding spine and/or leaf nodes does not change the topology
  • means adding capacity for service needs that change unpredictably, such as storage, is easy

Top of Rack Switches

• leaf layer access switches are implemented as top-of-rack switches

Top-of-rack (ToR) switches are switch models designed to provide high-speed connectivity to a rack of server appliances and support higher bandwidths than ordinary workgroup switches.

• may have 10 Gbps access port and 40/100 Gbps uplink ports
• usually place at the top of the rack
  • ensures cleaner cable management and better accessibility