Calculate switch port capacity requirements and oversubscription ratios.
A switch is 'non-blocking' when its internal fabric/backplane capacity is at least as large as the combined bandwidth demand of all ports running at full duplex simultaneously — meaning every port can run at line rate at the same time with zero contention. When fabric capacity is lower than total port demand, the switch is 'blocking' or oversubscribed at that ratio, meaning ports must share the internal fabric and won't all reach line rate simultaneously under full load. Oversubscription is a deliberate, common design choice for access-layer switches where all ports rarely peak simultaneously — but it becomes a real bottleneck if traffic patterns don't match that assumption.
Oversubscription ratio
ratio = (ports × speed × 2) / fabric_capacity
Not necessarily — access-layer/edge switches are commonly designed with 2:1 to 4:1 oversubscription because real-world traffic rarely has every port at 100% simultaneously. Oversubscription becomes a genuine bottleneck mainly for uplinks, aggregation-layer switches, or workloads (like storage networks) where sustained near-line-rate traffic on many ports at once is expected.
Core and spine switches are usually designed as non-blocking (1:1 or better) since they aggregate traffic from many access switches and a bottleneck there affects the whole network — access switches more commonly run 3:1 to 8:1 oversubscribed.
It's the rated port speed divided by the oversubscription ratio, representing the average bandwidth each port could sustain if all ports demanded traffic simultaneously and the fabric fairly shared its capacity — actual per-port throughput in practice depends on the switch's scheduling/QoS behavior and real traffic patterns.