What Is Latency? Meaning, Units and Causes Explained
Latency is the time it takes for data to travel from one point to another on a network, usually measured in milliseconds (ms). Most tools report round-trip time: the delay for a message to reach a server and for the reply to come back. It is set by distance, the quality of the route, the type of connection and how busy the equipment along the path is.
On this page
- What is latency, exactly?
- One-way latency vs round-trip time (RTT)
- What unit is latency measured in?
- What are the components of latency?
- The journey of one request
- Typical latency values by distance and network
- How latency is felt by users
- Latency vs lag vs ping
- How do you measure latency?
- Worked examples
- Common mistakes and myths
- What to do next
Quick answer: latency is the time a piece of data takes to get from one place to another on a network. It is measured in milliseconds (ms), where 1 ms is one thousandth of a second. Most tools report round-trip latency, the time for a request to reach a server and for the answer to come back. Lower is better, but what counts as low depends mostly on distance: a server in your own city can answer in under 10 ms, while a server on the opposite side of the planet cannot answer in less than roughly 150 to 200 ms because of the speed of light in fibre.
What is latency, exactly?
In networking, latency is the delay between a signal being sent and that signal being received or acted upon. It answers the question "how long until something starts to happen?" and not "how much can I move per second?" That second question is about bandwidth and throughput, and mixing the two up is the most common misunderstanding in this field.
Latency exists in every system that moves information. A web page load, a video call, a game input, a bank transfer and a DNS lookup all begin with a wait. If the wait is short you do not notice it. If it is long, the wait is multiplied by every step that needs a reply before the next step can start, which is why a modest-looking delay can make a page feel slow.
One-way latency vs round-trip time (RTT)
One-way latency is the time from sender to receiver. Round-trip time (RTT) is the time from the sender transmitting a message to receiving the reply. Almost every everyday tool measures RTT because it needs only one clock: the sender starts a timer, sends, and stops the timer when the reply arrives. Measuring true one-way delay requires two clocks synchronised to a fraction of a millisecond, which is difficult outside specialist setups.
It is tempting to say RTT = 2 × one-way, and it is often close, but not always. The outbound and return paths on the internet can be different routes (this is called asymmetric routing), so the two directions may have different delays. When people say "my latency is 40 ms" they almost always mean RTT, so about 20 ms each way on a symmetric path.
What unit is latency measured in?
Latency is measured in milliseconds (ms). Smaller units appear when talking about local networks and hardware: a microsecond (µs) is a thousandth of a millisecond, and local Ethernet switching delay is usually in microseconds. Larger values are sometimes written in seconds, for example a satellite connection with a 0.6 s round trip, but milliseconds remain the standard.
| Unit | Equals | Where you see it |
|---|---|---|
| 1 microsecond (µs) | 0.001 ms | Switch and NIC delay, light travelling 200 m in fibre |
| 1 millisecond (ms) | 0.001 s | Ping results, web and game latency |
| 100 ms | 0.1 s | Roughly the threshold where an interface starts to feel delayed |
| 1,000 ms | 1 s | A clearly slow response; a long wait for a click |
What are the components of latency?
Total network latency is the sum of four delays that apply at every hop along the path. Understanding them tells you which ones you can change and which ones physics fixes.
Propagation delay
The time a signal needs to physically travel through the medium. Light in optical fibre moves at about 200,000 km/s, roughly two thirds of its speed in a vacuum. That gives about 5 µs per km one way, or about 1 ms of round-trip time per 100 km of fibre. Nothing you buy can reduce this for a given route; only a shorter route or a closer server helps.
Transmission delay
The time to push all bits of a packet onto the wire: packet size divided by link speed. A 1,500-byte packet is 12,000 bits. On a 10 Mbps link that takes 1.2 ms; on a 1 Gbps link it takes 12 µs. This is where bandwidth does affect latency, and it matters mainly on slow links and for large packets.
Processing delay
The time each router, firewall, load balancer or server needs to examine the packet, look up where to send it, and, at the end, run application code. Router forwarding is typically microseconds. A web server building a dynamic page can add tens or hundreds of milliseconds.
Queuing delay
The time a packet waits in a buffer because the outgoing link is busy. It is zero on an idle link and can grow to hundreds of milliseconds on a congested one. Queuing is the most variable component, and the main cause of jitter and of bufferbloat.
| Component | Depends on | Can you improve it? |
|---|---|---|
| Propagation | Path length, medium | Only by choosing a closer server or a more direct route |
| Transmission | Packet size, link speed | Yes, on slow links, by upgrading the link |
| Processing | Hardware, software, server load | Yes, mostly on the server side |
| Queuing | Congestion, buffer size | Often, by reducing load or using smart queue management |
The journey of one request
Consider you opening a page on a server in another city. Even before any content arrives, several round trips happen in sequence:
- DNS lookup: your device asks a resolver for the server's IP address (see DNS explained). Cached answers cost almost nothing; an uncached one can cost one or more round trips.
- TCP handshake: one round trip to open the connection (see TCP and UDP).
- TLS handshake: for HTTPS, typically one more round trip with TLS 1.3, two with TLS 1.2.
- HTTP request and response: one more round trip plus the time the server spends generating the answer.
With a 50 ms RTT, a fresh HTTPS connection to a server can spend roughly 50 ms on TCP, 50 ms on TLS 1.3, and 50 ms on the first HTTP request, so about 150 ms before the first byte of content arrives, excluding DNS and server time. That is why each millisecond of RTT is paid for several times. The same request on an already open connection costs just one RTT.
Typical latency values by distance and network
The figures below are physical minimums for fibre plus realistic typical ranges. Real routes are longer than straight lines because cables follow roads, railways and seabeds, and traffic passes through several networks. Treat the ranges as rough guides, not guarantees.
| Scenario | Physical minimum RTT | Typical real-world RTT |
|---|---|---|
| Same building (wired LAN) | Well under 1 ms | Under 1 ms |
| Same city | Under 1 ms | Roughly 5 to 20 ms |
| Same country (about 1,000 km) | About 10 ms | Roughly 15 to 40 ms |
| Across a continent (about 4,000 km) | About 40 ms | Roughly 50 to 100 ms |
| Transatlantic (about 5,600 km) | About 56 ms | Roughly 70 to 100 ms |
| Europe to east Asia or Australia | About 150 ms or more | Roughly 200 to 300 ms |
| Geostationary satellite link | About 480 ms (orbit height alone) | Often 550 to 700 ms |
The access technology adds its own share on top: wired Ethernet adds almost nothing, fibre-to-the-home a few milliseconds, cable and DSL often 5 to 20 ms, Wi-Fi a few milliseconds that can vary, and mobile networks commonly 20 to 60 ms depending on generation and signal. See Wi-Fi, Ethernet, fibre, mobile and satellite latency and undersea cables and the speed of light for details.
How latency is felt by users
People do not perceive milliseconds directly; they perceive consequences.
- Web browsing: each new connection costs several RTTs, so pages from far-away servers start more slowly. Differences below roughly 100 ms are hard to notice on their own; pages that need dozens of sequential requests amplify them.
- Video calls: conversation becomes awkward when one-way delay exceeds roughly 150 ms, as people start talking over each other.
- Online games: actions are processed on the server, so your input is delayed by the RTT. Fast shooters and fighting games punish tens of milliseconds; strategy games tolerate far more.
- Remote desktop and SSH: every keystroke waits for the echo, so high latency feels like typing through mud.
- Trading and industrial control: differences of microseconds can matter, which is why firms place servers physically next to exchanges.
See what is a good latency for tables by use case.
Latency vs lag vs ping
| Term | What it means | Measured in |
|---|---|---|
| Latency | The general concept: delay between cause and effect on a network | ms |
| Round-trip time (RTT) | Latency to the destination and back | ms |
| Ping | A specific tool using ICMP echo messages, also used as slang for RTT | ms |
| Lag | The experience of delay or stutter, from any cause | Not a unit; subjective |
| Jitter | Variation in latency from one packet to the next | ms |
Not all "ping" numbers are the same measurement. ICMP ping and HTTPS timing can differ for the same server, as explained in ping vs HTTPS latency.
How do you measure latency?
Command-line tools
ping example.com sends ICMP echo requests and prints the RTT of each. traceroute (Linux, macOS) or tracert (Windows) shows the hops on the way and the delay at each, and mtr combines both continuously. These show network-layer timing but some networks block or deprioritise ICMP.
Browser-based timing
A browser can time real HTTPS requests without installing anything. GlobeLatency does this: it makes HTTPS requests from your own device to public servers around the world and times each one from the start of the request until the response headers arrive. It is not ICMP ping, so its numbers include DNS, TCP and TLS setup when a new connection is needed, and are therefore a bit higher than raw ICMP. You can run it once, or watch destinations continuously on the live map or the monitor.
Take several samples
One reading is almost meaningless because latency fluctuates. Look at the average, the best and worst values, and ideally a high percentile such as the 95th. See mean, median, min and max and percentiles.
Worked examples
Example 1: Why is a server 6,000 km away never below about 60 ms?
At about 5 µs per km one way, 6,000 km of straight fibre gives 30 ms one way, or 60 ms RTT. Real cable paths are longer, say 7,500 km of fibre, giving 37.5 ms one way and 75 ms RTT. Add roughly 5 to 10 ms for equipment and access networks and a measurement of about 80 to 85 ms is excellent, not poor.
Example 2: Adding up a request
You are 30 ms RTT from a web server. DNS answer cached (0 ms), TCP handshake 30 ms, TLS 1.3 handshake 30 ms, HTTP request 30 ms of network plus 40 ms of server processing. Time to first byte is 30 + 30 + 30 + 40 = 130 ms. Moving to a server 10 ms away would cut it to 10 + 10 + 10 + 40 = 70 ms. The server's 40 ms was unchanged, but the network share fell by two thirds.
Example 3: Transmission delay on a slow link
A 1,500-byte packet is 12,000 bits. At 1 Mbps it takes 12 ms to transmit; at 100 Mbps, 0.12 ms. On a slow link, a few queued full-size packets ahead of yours add tens of milliseconds, which is why a saturated upload can wreck responsiveness.
Common mistakes and myths
- "More Mbps means lower ping." Bandwidth and latency are separate. A faster plan only helps latency when the old link was saturated.
- "My ping to the other side of the world should be 20 ms." Physics forbids it. Compare against the distance, not against a universal number.
- Judging from one test. A single sample can be an outlier. Repeat the test and look at the spread.
- Comparing numbers from different tools directly. ICMP ping, TCP connect time and HTTPS timing measure different things.
- Testing over Wi-Fi and blaming the ISP. Compare a wired test to separate home wireless from the provider.
- Forgetting that latency to a destination depends on the destination. Your connection can be perfect and a particular server can still be slow or far away.
What to do next
Run a quick check with the free latency test, then read what is a good latency to interpret the numbers for your use. If your results look high, troubleshoot high latency walks through isolating the cause, and how to reduce latency lists what actually helps. To understand why bandwidth is a different matter, continue with latency vs bandwidth vs throughput.
Frequently asked questions
What does latency mean in simple terms?
Latency is the delay between doing something online and seeing the result begin to happen. Technically it is the time data needs to travel from a source to a destination. In everyday use, people usually mean round-trip latency: how long it takes for a request to reach a server and for the first part of the answer to return, measured in milliseconds.
Is lower latency always better?
Yes, for the same path and the same measurement method, lower is better, because every interaction waits on it. But there is a physical floor set by distance, so a very low number is only possible to nearby servers. A 20 ms result to a server 8,000 km away would be impossible; a 20 ms result to a server in your city is ordinary.
Is latency the same as ping?
Not exactly. Ping is a specific tool (and the ICMP echo message it sends) that reports round-trip time. Latency is the general concept of delay. People often use "ping" casually to mean round-trip latency, especially in gaming, but different tools measure at different layers and can give different numbers for the same server.
What is the difference between latency and lag?
Latency is a measurable network delay in milliseconds. Lag is what a person experiences: stutter, rubber-banding, delayed button presses. Lag can be caused by high latency, but also by packet loss, jitter, a slow computer, low frame rates or an overloaded server. High latency does not always mean lag, and lag does not always mean high latency.
How is latency measured?
By timing a message from the moment it is sent until a reply (or the first byte of a reply) arrives. Common methods are ICMP ping, traceroute or mtr, TCP connect timing, and timing real HTTPS requests in a browser. The result is expressed in milliseconds. Taking several samples and looking at the average, best and worst values gives a more reliable picture than a single reading.
Can latency be zero?
No. Signals travel at a finite speed, about 200,000 km/s in optical fibre, and every router, switch and server adds processing time. Even to a machine in the next room the round trip takes a fraction of a millisecond, and across oceans it takes tens or hundreds of milliseconds no matter how fast the connection is.
Does a faster internet plan reduce latency?
Usually not by much. Higher bandwidth lets you move more data per second but does not shorten the distance or the number of hops. Moving from a 100 Mbps plan to a 1 Gbps plan on the same technology typically leaves latency to distant servers almost unchanged, although it can help if your old line was congested.