DATA CENTER · SWITCHING · FABRIC
Plan Leaf-Spine and three-tier data center convergence, east-west/north-south traffic, Spine N-1 resilience, bottlenecks and growth.
Set the engineering assumptions and design parameters.
Review calculated capacity, limits and design checks.
| Year | Demand | Fabric | Upper | Egress | Status |
|---|
Layer 1 compares IT-facing line rate with fabric capacity, layer 2 compares fabric with the upper tier, and end-to-end compares IT-facing line rate with final egress.
The model separates traffic, counts two fabric-facing traversals for cross-edge east-west flows, and lets you define the share that continues through an upper tier.
Assumes edge uplinks are evenly distributed across middle-layer nodes and reduces fabric and upper capacity by the failed-node share.
Validate with P95/P99 utilization, queue drops, ECMP/LAG distribution, ASIC limits and failure testing before production changes.
RFC 7938 · RFC 7424 · IEEE 802.1AX
Content reviewed: 2026-09-11
It is the ratio between downstream capacity offered by servers or access switches and upstream capacity at the next layer. A 3:1 ratio means three units of downstream line-rate capacity share one unit of upstream capacity.
They are closely related. Oversubscription often describes one switch boundary; convergence ratio is useful for capacity narrowing across multiple fabric layers.
The model assumes edge uplinks are evenly distributed across Spine or aggregation nodes. Removing one middle-layer node reduces fabric and upper-layer capacity by that node share.
A cross-Leaf flow consumes a fabric-facing link on the source Leaf and another on the destination Leaf, so both traversals are counted for aggregate fabric-port screening.
No. Validate with P95/P99 utilization, ECMP distribution, queue drops, flow telemetry, forwarding limits and failure testing.
ENGINEERING WORKFLOW
Move from traffic sizing through switch uplinks, fabric convergence, optics and implementation planning for a data center network.
