Reducing network latency: where it comes from and how to control it

Latency and bandwidth are often confused. Bandwidth is how much data can be carried per second; latency is how long a packet takes to travel from A to B. For anything interactive — voice, video conferencing, online gaming, trading, responsive web apps — latency shapes the experience far more than raw bandwidth.

Where does latency come from?

• Physical distance: light in fibre travels no faster than ~200,000 km/s. Paris–New York is irreducibly a few tens of milliseconds round trip.
• Hop count and route quality: every router crossed adds a little delay; a long or detoured BGP path adds a lot.
• Congestion: a saturated link queues packets — "bufferbloat", which makes latency spike under load.
• Processing: mitigation, inspection, undersized equipment can all add delay.

The levers to reduce it

Bringing content closer to users (CDN, multiple points of presence, anycast) shortens distance. Quality routing — rich peering, presence at IXPs, short paths — avoids needless detours. Sufficient capacity prevents congestion and therefore queuing. Finally, well-tuned QoS prioritises sensitive traffic (voice, video) over background traffic (backups).

How to measure it honestly

A simple ping gives the RTT (round trip), but also watch jitter (latency variation, critical for voice) and latency under load, not just at rest. A traceroute reveals the path and the hops where delay accumulates. A Looking Glass lets you observe routes from the operator's own network.

The operator's role

Latency is largely decided in the network you don't see: interconnection quality, route length, absence of congestion. An operator running its own richly interconnected, over-provisioned core delivers your packets over the shortest path. That's what our IP transit is about; you can also follow the health of our links on our status page.