Networking in the Data Center
Data center networking covers topology, addressing, and the physical layer. Know the layered model (access/leaf, aggregation/spine, core), the idea of leaf-spine for predictable east-west latency, and the roles of top-of-rack (ToR), end-of-row (EoR), and aggregation switches. Understand VLANs for segmentation, trunk versus access ports, link aggregation for bandwidth and redundancy, and spanning tree or its modern replacements for loop prevention.
Key concepts: the difference between bandwidth and latency; oversubscription ratios (how much access bandwidth is backed by uplink bandwidth); and redundancy protocols such as LACP, MLAG, and routing-based designs. Know that a single-homed server is a single point of failure regardless of switch redundancy. Cabling matters: copper for short runs, fiber for distance and EMI immunity, and transceiver/optics compatibility.
Traps: assuming link aggregation increases single-flow throughput (it usually balances flows, not one flow); confusing a VLAN with a subnet (they often align but are distinct); and forgetting that a redundant pair of switches only helps if the server dual-homes to both.
Worked example / rule of thumb: a ToR switch with 48 × 10 GbE downlinks and 4 × 40 GbE uplinks has 480 Gb down and 160 Gb up, roughly 3:1 oversubscription. Rule of thumb: design for the traffic pattern (north-south vs east-west), and remember redundancy requires two independent paths end to end.