throughput

Latency fine, loss low, link idle, yet a single copy takes hours. The arithmetic of bandwidth-delay product settles this question before any tool opens.

Most throughput tests measure the tester, not the network. Single stream versus parallel, slow start, sender-side retransmits, and the CPU ceiling: how to produce a number you can act on.

Both configurations fit the weights. Compare cache capacity, collective overhead and replica placement under the actual traffic shape; neither layout wins every workload.

It trades compute for latency, so it wins on an underloaded fleet and can lose on a saturated one. The acceptance rate decides everything, and the acceptance rate depends on traffic you do not control.

Expensive accelerators waiting on cheap storage is the most common form of waste in a training cluster, and it looks exactly like a code problem on every dashboard.

Nobody is waiting, which changes every constraint. Latency stops mattering, throughput per dollar becomes the only metric, and the design converges on something that looks like a data pipeline rather than a service.