Size switch or router buffers using the BDP rule or the Stanford BDP over root-N rule for many flows.
One flow needs a full BDP of buffer to keep the link busy through its sawtooth. With many independent flows the sawtooths desynchronise and their aggregate is far smoother, so the buffer requirement falls by the square root of the flow count. This is why datacentre switches ship with shallow buffers and still run at line rate: with hundreds of flows per port, buffer memory sized to a full BDP would only add latency.
Buffer Size
buffer = BDP for a single flow, or BDP ÷ √N when N long-lived flows share the link.
buffer = BDP for a single flow, or BDP ÷ √N when N long-lived flows share the link. One flow needs a full BDP of buffer to keep the link busy through its sawtooth. With many independent flows the sawtooths desynchronise and their aggregate is far smoother, so the buffer requirement falls by the square root of the flow count.
This is why datacentre switches ship with shallow buffers and still run at line rate: with hundreds of flows per port, buffer memory sized to a full BDP would only add latency.
This calculator takes 4 inputs: Link speed, Round-trip time, Concurrent long-lived TCP flows, Use the BDP ÷ √N rule. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
Use the full BDP when a small number of high-rate flows dominate the link — bulk replication, backup windows or a single large file transfer. The √N reduction relies on many independent flows whose congestion cycles do not line up.