Latency is the time it takes for data to travel from one point on a network to another, usually measured in milliseconds (ms). It is often reported as round-trip time, the time for a packet to reach its destination and for the reply to come back. Low latency makes calls, video meetings and interactive applications feel responsive; high latency makes them feel delayed, even when the connection has plenty of bandwidth.
At a glance
- Latency is delay, not capacity: how long data takes to arrive, not how much can be sent at once.
- Distance sets a floor, because signals in fiber travel at a finite speed; routing, queuing and processing add more on top.
- It is quoted either one-way or round-trip, and the two differ by roughly a factor of two, so check which a figure means.
- Real-time traffic such as voice, video and remote desktop is usually the most sensitive to it.
- Variation in latency is called jitter, and is often as important as the average.
What problem it solves
Latency is not something a buyer wants; it is something to keep low and predictable. Understanding it helps explain performance problems that a speed test does not show. A conversation where people keep interrupting each other, a cloud application that pauses after every click, or a remote desktop that lags behind the mouse are usually latency problems, not bandwidth problems.
Knowing where latency comes from lets a buyer choose the right fix. If the cause is distance, the answer may be a closer cloud region, a better route or a direct connection. If it is congestion, more capacity or traffic prioritization helps. If it is a detour through a distant data center or security service, the answer is a design change.
How it works
Total latency is the sum of several delays along the path:
Propagation. The time a signal takes to travel the physical distance. Light in fiber covers roughly 200 kilometers per millisecond, so a cross-country or intercontinental path has a minimum delay no service can remove. Real fiber routes are longer than the straight-line distance.
Serialization and processing. Each router, firewall and device adds a small delay to receive, inspect and forward each packet. Security inspection and encryption can add more.
Queuing. When a link is busy, packets wait in buffers before being sent. This is the part that grows with congestion and changes from moment to moment, which is why latency is often worse at peak hours. Quality of Service (QoS) settings can let time-sensitive packets skip the queue on links you control.
Path choice. Internet routing chooses paths based on agreements between networks, not always the shortest route. Two providers can reach the same destination with different delays. SD-WAN tools measure latency on each available link and can steer sensitive applications onto the better one.
Latency is usually measured with tools such as ping or traceroute, by monitoring built into network equipment, or by the applications themselves. Together with jitter and packet loss, it largely determines the quality of experience for real-time applications.
When it matters for buyers
- Rolling out voice or video. Calling and meeting platforms are sensitive to delay, so measure latency from each site to the provider before cutover.
- Moving applications to the cloud. An application that worked well on the local network may slow down when every request crosses the internet. Region choice and private connectivity can help; see our cloud connect page for options.
- Adding sites far from the data center or cloud region. Distance alone may make some applications unworkable without redesign.
- Choosing between providers. Ask for measured latency from your sites to the services you use, not a generic backbone figure.
- Adding security in the path. Routing traffic through a cloud security service or central firewall can add delay, depending on where its points of presence are.
Questions to ask vendors
- What latency should we expect from our sites to the cloud regions and services we use, and how was that measured?
- Does your SLA include a latency target? Between which points, measured how, and averaged over what period?
- Where does our traffic reach the internet or your backbone, and where are your nearest points of presence?
- How do you handle congestion at peak hours, and do you prioritize real-time traffic?
- Can we see latency reporting for our circuits or sites?
- If performance degrades, can your tools show which segment of the path added the delay?
How it differs from bandwidth
Bandwidth is how much data a connection can carry each second; latency is how long data takes to arrive. A connection can have high bandwidth and high latency, as with many satellite services, or modest bandwidth and low latency. Downloads and backups depend mainly on bandwidth, while calls and interactive applications often depend more on latency. Adding bandwidth reduces latency only when the existing link is congested and packets are waiting in queues.