Traditional data center networks are often built using a layered three-tier design with access, aggregation, and core. These designs frequently rely on Layer 2 domains and spanning tree to prevent loops. One major drawback is inefficient resource usage. Spanning tree typically blocks redundant links, which means installed bandwidth and expensive uplinks can sit idle during normal operation. To compensate, organizations overbuild capacity or use complex workarounds, raising cost and operational burden. In addition, scaling tends to be vertical, adding bigger chassis or more complex aggregation, which can concentrate risk and reduce flexibility.
Another common challenge is increased latency. In a three-tier model, traffic between servers in different access blocks often traverses multiple tiers, adding hops and delay. Even when traffic stays within the same facility, the path is not consistently short, and application performance can suffer, especially for east-west workloads such as distributed databases and microservices where frequent server-to-server communication is normal. Modern leaf spine IP fabrics reduce these issues by using a routed underlay with predictable one-spine-hop paths between leaves and by keeping all links active with multipath routing.
While traditional architectures can also experience suboptimal path behavior, the most consistently cited challenges are idle capacity due to blocked links and additional hop count that increases latency.
