Which capability is provided by a spine-and-leaf architecture?

A spine-and-leaf architecture provides predictable latency because all traffic paths between any two leaf switches are exactly two hops (leaf-to-spine, then spine-to-leaf). This consistent path length eliminates the variable latency found in traditional three-tier hierarchical networks, where paths can vary in length depending on the location of the source and destination. The option "Only sends multicast traffic between servers directly connected to the spine" is incorrect because spine-and-leaf architectures handle all types of traffic, and multicast is not restricted to spine-connected servers. "Reduces oversubscription by adding a leaf layer" is incorrect; while spine-and-leaf designs aim to reduce oversubscription, it's achieved through high-bandwidth links and proper sizing, not simply by adding a leaf layer. "Restricts payload size within the leaf layer" is incorrect; payload size is typically determined by network protocols and MTU settings, not by the architecture itself.