Latency is the time delay between sending a request and receiving a response. It is measured in milliseconds and affects every user interaction on the internet.
TL;DR Latency is the time it takes for data to travel from a sender to a receiver and back, measured in milliseconds (ms). In networking, latency is often expressed as RTT (Round-Trip Time). Typical broadband internet latency is 10–50ms to nearby servers and 100–300ms to servers on other continents. Latency above 100ms is noticeable to users in interactive applications. Latency above 300ms is considered poor. The four main causes of latency are propagation delay (physical distance), transmission delay (bandwidth), processing delay (hardware/software), and queuing delay (congestion). Edge computing reduces latency by processing requests closer to users instead of routing them to centralized data centers.
What is latency?
Latency is the elapsed time between a request being sent and a response being received. In networking, it is most commonly measured as RTT (Round-Trip Time) — the time for a packet to travel from the client to the server and back.
Latency is measured in milliseconds (ms). Lower latency means faster responses.
Latency is not the same as bandwidth. Bandwidth is how much data can be transferred per second. Latency is the delay before that transfer begins. A high-bandwidth connection can still have high latency.
The 4 components of network latency
Network latency is the sum of four distinct delays:
| Component | Definition | Controllable? |
|---|---|---|
| Propagation delay | Time for a signal to travel the physical distance between sender and receiver. Light travels through fiber at ~200,000 km/s. | No — bounded by physics |
| Transmission delay | Time to push all packet bits onto the wire, dependent on link bandwidth. | Partly — improve bandwidth |
| Processing delay | Time for routers, switches, and servers to inspect and process packet headers. | Yes — hardware and software optimization |
| Queuing delay | Time packets wait in buffers under network congestion. | Yes — traffic management, capacity planning |
Queuing delay is the most variable component — it can spike from near-zero to hundreds of milliseconds under congestion.
Typical latency values by scenario
| Scenario | Typical latency |
|---|---|
| Same data center (local network) | < 1 ms |
| City to city, same country | 5–20 ms |
| Cross-continent (e.g., US East to US West) | 60–80 ms |
| Transatlantic (US to Europe) | 80–120 ms |
| Transpacific (US to Asia) | 150–300 ms |
| 4G mobile network | 30–100 ms |
| 5G mobile network | 1–10 ms |
| Satellite internet (LEO, e.g., Starlink) | 20–60 ms |
| Satellite internet (GEO) | 500–700 ms |
| Wired LAN (Ethernet) | < 1 ms |
A server in São Paulo delivers responses in ~15ms to users in São Paulo and ~170ms to users in Tokyo — not because of server speed, but because of physical distance.
Latency vs bandwidth
| Latency | Bandwidth | |
|---|---|---|
| Definition | Delay before data transfer begins | Maximum data throughput per second |
| Measured in | Milliseconds (ms) | Megabits per second (Mbps) or Gbps |
| Analogy | How long before the first truck leaves | How many trucks can travel per hour |
| Affected by | Distance, routing, congestion, processing | Link capacity, hardware, protocol |
| Improves with | Edge computing, optimized routing, caching | Upgrading network links |
High bandwidth doesn’t fix high latency. A 10 Gbps connection with 300ms latency still feels slow for interactive applications.
Latency impact by application type
| Application | Acceptable latency | Poor latency threshold | Why it matters |
|---|---|---|---|
| Web page load (TTFB) | < 200ms | > 500ms | Google: 53% of mobile users abandon pages taking over 3 seconds |
| API calls | < 50ms | > 200ms | At 1,000 calls/sec, 1ms extra latency = 1s of added overhead |
| Video streaming | < 150ms | > 500ms | Causes buffering and playback interruptions |
| Online gaming | < 50ms | > 100ms | Noticeable “lag” that affects gameplay |
| VoIP / video calls | < 150ms | > 300ms | Causes echoes, talk-overs, degraded call quality |
| High-frequency trading | < 1ms | > 5ms | Microsecond differences determine profit and loss |
| Real-time AI inference | < 100ms | > 500ms | Determines whether AI-powered interfaces feel responsive |
How to reduce latency
The most effective latency reduction strategies, ordered by impact:
- Bring compute closer to users — Processing requests at servers near the user eliminates transcontinental round-trips. A user in Jakarta served from a nearby distributed node gets 5–15ms latency instead of 150ms from a US-based origin.
- Cache content at the network edge — Serve static and semi-static content from distributed cache nodes, eliminating origin round-trips for cached assets entirely.
- Use HTTP/3 and QUIC — QUIC eliminates TCP connection setup overhead and handles packet loss without head-of-line blocking, reducing latency by 10–30% on mobile networks.
- Terminate TLS close to users — TLS handshakes require 1–2 round-trips. Terminating TLS at a distributed node near the user reduces handshake latency from 200ms (to a distant origin) to under 10ms.
- Minimize DNS lookup time — Use anycast DNS with global points of presence. Slow DNS adds 50–200ms to every new connection.
- Reduce payload size — Compress responses with Brotli or gzip. Smaller payloads transmit faster over the same link.
- Reuse connections (keep-alive) — Reusing established TCP and TLS connections eliminates handshake latency for subsequent requests to the same server.
Frequently asked questions
What is latency in networking? Latency in networking is the time it takes for a data packet to travel from one point to another and back, measured in milliseconds. It is commonly expressed as RTT (Round-Trip Time). Lower latency means faster response times.
What is the difference between latency and bandwidth? Latency is the delay before data transfer starts — how long you wait before anything arrives. Bandwidth is how much data can be transferred per second — how much you get once it starts. A high-bandwidth connection can still have high latency. Both matter for performance but are independent of each other.
What is RTT? RTT (Round-Trip Time) is the time it takes for a signal to travel from a sender to a receiver and back. RTT is the most common way to measure network latency. The ping command measures RTT.
What causes high latency? The most common causes of high latency are: (1) physical distance between the client and server, (2) network congestion causing queuing delays, (3) too many network hops between origin and destination, and (4) slow server processing times.
What is a good latency for a website? Time to First Byte (TTFB) below 200ms is considered good. TTFB above 500ms is poor. Google’s Core Web Vitals recommend TTFB under 800ms for a passing score, but under 200ms for optimal user experience.
What is the difference between latency and ping? Ping is a command-line tool that measures RTT by sending ICMP echo packets to a destination and timing the response. Latency is the concept being measured. Ping is one way to measure it.
What is p95 or p99 latency? P95 (95th percentile) latency means 95% of requests complete faster than this time. P99 means 99% complete faster. Median (p50) latency can look excellent while p95/p99 are terrible — those tail latencies represent real users having poor experiences. For production systems, monitor p95 and p99, not just averages.
Does edge computing reduce latency? Yes. Edge computing reduces latency by processing requests at servers geographically close to users instead of routing all traffic to centralized data centers. A user in São Paulo served from a distributed node in São Paulo experiences 5–15ms latency instead of 150–300ms to a distant origin server.