Two Different Measures
People conflate speed, but networks have two distinct properties. Latency is the time for one unit of data to travel from sender to receiver. Throughput is the volume of data delivered per unit of time.
Why They Are Independent
A satellite link can have huge throughput yet terrible latency, while a short copper cable has low latency but limited capacity. Improving one does not automatically improve the other.
- Latency is bounded by distance and the speed of light, plus queuing and processing delays.
- Throughput is bounded by bandwidth and by protocol behavior such as the TCP window.
The Bandwidth Delay Product
The bandwidth delay product is bandwidth multiplied by round trip latency. It tells you how much data must be in flight to keep a high latency link full. If the TCP window is smaller than this product, throughput stays low even on a fat pipe.
Practical Effects
- Many small requests are dominated by latency, so batching and connection reuse help.
- Bulk transfers are dominated by throughput, so larger windows and parallelism help.
Optimizing the wrong one wastes effort, so always measure which dominates your workload.
Key idea
Latency is the delay of one trip and throughput is data volume per time, and the bandwidth delay product links them when sizing in flight data.