Contents

Web & Networking › Networking Fundamentals

Throughput

How much work or data gets through per unit of time.

Also known as: throughput, bandwidth, bandwidth vs throughput

Throughput is the rate data actually moves — bits per second achieved — while bandwidth is the capacity of the link. People conflate them constantly, and worse, conflate both with speed (how fast any one interaction feels). A gigabit link doesn’t make a chatty protocol fast; it just moves big transfers quickly once they start.

bandwidth:   the pipe's width (capacity)
throughput:  what you actually get (limited by loss, RTT, protocol chatter)
latency:     how long one interaction takes (round trips × RTT)

Throughput is capped by the slowest of: link capacity, packet loss (TCP backoff punishes loss brutally), round-trip count for the transfer (handshakes, request/response turns), and either endpoint’s processing. Small files over long links are latency-bound; only large sustained transfers are bandwidth-bound.

The classic mistakes:

  • Upgrading bandwidth to fix latency. Interactive slowness from round trips doesn’t budge with a fatter pipe. Count round trips first.
  • Measuring with one big file. A speed test shows capacity, not what your API’s hundred tiny requests experience. Measure with representative traffic.
  • Ignoring loss. Even 1% loss can halve TCP throughput via congestion control. Throughput problems are often loss problems.
  • Uncompressed or unbatched transfers. Sending verbose JSON per record wastes the throughput you have; compression and batching multiply effective throughput for free.
  • Single-connection limits. One TCP connection ramps slowly (slow start) and stalls on loss; parallel connections or modern transports raise sustained throughput.
  • Forgetting the endpoints. Disk, TLS, serialisation and application logic all cap throughput before the network does. Profile end to end.
  • Quoting bits vs bytes. Link speeds are bits; file sizes are bytes. An 8× “discrepancy” is just units.

The discipline: latency is round trips × RTT; throughput is the minimum of capacity, loss-limited rate, and endpoint speed. Optimise the binding constraint — usually round trips for interactive use, and transfer efficiency for bulk.