Cloud latency explained: how region and network affect performance
Latency is the delay in a connection; distance to the region and the number of network hops determine how fast a service feels.
Latency is the time between a request and the start of the response. In the cloud, it depends on the distance to the region, the network route, the number of steps in your architecture and the load. Choose a region close to your users and measure latency from their situation.
Who is it for: Developers, DevOps engineers and technical buyers who want to explain and improve the performance of cloud applications.
What is latency and how does it differ from bandwidth?
Latency is the waiting time of a signal from A to B and back, often expressed as round-trip time (RTT). Bandwidth is how much data per second fits through a connection. You can have plenty of bandwidth and still experience sluggishness if every action has to go back and forth several times.
The difference matters for small, frequently repeated requests, such as API calls, database queries and payment flows. With large files, bandwidth matters more; with interactive applications, latency determines the experience.
What influences latency?
Distance is the lower limit: signals don’t travel faster than the medium allows. On top of distance come the network route, the number of intermediate hops, the load on connections and the processing time of the service itself.
Architecture counts for a lot. A page that makes five consecutive calls to another region experiences that delay five times over. Setting up an encrypted connection also takes a few round trips, which makes reusing connections important.
How do you choose a region and optimise?
Place the service close to the largest group of users, but take data requirements into account: close by is not allowed or desirable in every case. Keep components that talk to each other often, such as application and database, in the same zone.
- Put static content behind a CDN or cache
- Avoid consecutive requests to distant sources; bundle them or work in parallel
- Reuse connections and use modern protocols
- Use multiple regions if users are widely spread
- Measure with real users, not only from your own office
How do you measure and monitor latency?
Don’t look only at the average. Percentiles, such as the 95th and 99th percentile, show what the slowest group experiences. Measure from multiple locations and set a target per critical action, so you notice deterioration before users complain.
Causes of latency
| Factor | What happens | What can you do |
|---|---|---|
| Distance | Signal needs travel time to the region | Choose a region close to users |
| Network route | More intermediate hops add time | Ask the provider about routes and peering |
| Number of requests | Every call counts again | Bundle, work in parallel, cache |
| Connection set-up | Handshakes cost round-trip traffic | Reuse connections |
| Load | Overloaded components respond more slowly | Scale and monitor queues |
Who is it less suited for?
- Latency optimisation is less relevant for batch processing, where throughput matters more than response time.
- Multiple regions are less suitable when consistent data storage in one place is mandatory.
- A CDN helps little with highly personalised, non-cacheable responses; optimise the underlying chain instead.
Frequently asked questions
What is good latency?
That depends on the application. Real-time applications require much stricter values than a reporting tool. Set your own target per action.
What is jitter?
Jitter is the variation in latency. For voice, video and games it is often more disruptive than a slightly higher fixed value.
Will a faster internet connection help me?
Only if bandwidth is the bottleneck. For latency, only the route, distance or the number of requests changes anything.
Can I reduce latency between regions?
Not below the physical minimum, but you can make fewer requests between regions, replicate data or cache it.
