Bandwidth and latency

Bandwidth and latency are the two numbers that describe how fast a network connection is. Bandwidth is how much data the connection can carry each second, measured in megabits or gigabits per second (Mbps or Gbps). Latency is how long one piece of data takes to reach the other end, measured in milliseconds. The "100 Mbps" in a broadband plan is its bandwidth. The "ping" shown in an online game is its latency, usually measured there and back.

Here is how the two work together. The sender splits data into small pieces called packets. Bandwidth decides how many packets can leave each second. Latency is the time each packet takes to travel. It grows with distance, because light in fibre covers about 200 kilometres every millisecond. Each router on the way adds a little more delay. And when a link is full, packets wait in a queue, which adds more. So the time to receive something is roughly the latency plus its size divided by the bandwidth.

This is why the two matter for different jobs. A 100 MB file on a 100 Mbps line takes about 8 seconds, and a delay of 20 milliseconds hardly changes that. But a web page that makes 30 small requests one after another, each taking 100 milliseconds there and back, needs 3 seconds even on a very fast line. Adding bandwidth does not reduce latency. Only a shorter path, or fewer round trips, does that. Edge computing shortens the path by running servers close to the users.

Take, for example, an internet connection by satellite. A geostationary satellite is about 36,000 kilometres up, so a round trip takes about 600 milliseconds, while a low-orbit satellite needs only a few tens of milliseconds.

Consider a video streaming app. It needs steady bandwidth, so a content delivery network keeps copies of each video in servers near its viewers, which also cuts latency.

Inside a data centre, for instance, thousands of GPUs training one AI model swap results many times a second. AI cluster networking gives them links of hundreds of gigabits per second and delays of a few microseconds.

How are bandwidth and latency different from throughput? Throughput is the amount of data that actually arrives each second. It is usually lower than the bandwidth, because of lost packets, queues and the time spent waiting for replies. Mobile app performance testing measures this real throughput on real phone networks.