Traditional data center network fabrics are often designed with three tiers of switches that are core, aggregation, and access. Access switches connect to aggregation switches, which in turn connect to the core switches. The design topology of the physical layer can impact the efficiency and latencies.
Communication between two endpoints within the data center begins from the access switch, traverses the aggregation switch to the core switch, and travels from there to the remote endpoint. This traffic pattern results in inefficiencies and increased latency. While adequate for basic network traffic, this is not a reasonable traffic pattern for traffic that does not tolerate latency.
A two-tier leaf-and-spine network architecture is the more preferred approach for building new data center infrastructure. The two-tier architecture uses an access switch, or leaf, which is connected to an aggregation switch, or spine. The leaf switch provides connectivity between endpoints in the data center, while the spine switch provides high-speed interconnectivity between leaf switches. The leaf-and-spine network is connected in a full mesh, providing predictable communication and latency between endpoints. Ethernet connectivity is used from the host to the leaf switch, and the broadcast domain terminates at the leaf. External Border Gateway Protocol (eBGP) is the control plane option for routing within the leaf-and-spine architecture.
