A data-centre choice affects response time, connection stability and operational access. Geography matters, but carrier connectivity and routing determine the path traffic takes.
Run the checks before moving data and save the results so you can compare them after launch.
Start with a map of users
Identify where the main audience, staff and connected systems are located. A public website should prioritise visitors; an internal service should include offices and remote teams; an API should include the servers and cloud services that call it.
One test point is not enough for a distributed audience. Use at least a primary network, a distant one and a mobile connection.
NextMeasure more than average latency→Measure more than average latency
Average latency tells only part of the story. Variations are noticeable in calls, games and interactive applications, while a small but constant packet loss can damage an otherwise fast route.
- average latency and its variation;
- packet loss over a longer series;
- the route in both directions when possible;
- download and upload speed for a large file;
- results at different times of day.
How to use a test IP
A test IP helps assess the route to a facility, but it does not guarantee an identical path to the final server. Confirm that the address belongs to the same network and location as the intended service.
Run several latency series, save traceroute or mtr output and repeat the check in the evening. One good result does not show how the network behaves under load.
NextAssess more than the network→Assess more than the network
Power, cooling, physical access, connectivity and incident procedures all matter. A provider handles some of this for a rented server, but hardware replacement terms and support access still need to be clear.
- how power and connectivity are made redundant;
- who replaces failed drives or memory and how;
- whether remote console and power control are available;
- traffic and port restrictions;
- how maintenance and incidents are communicated.
When one location is not enough
A critical service may need a standby system in another facility. A second server in the same rack does not cover every failure. The greater the cost of downtime, the more independent the locations should be.
A standby system is still useless without regular synchronisation, switch-over tests and a named owner for the procedure. Design recovery first, then choose the second location.
NextA short pre-order checklist→A short pre-order checklist
Record the city and network, test IP, measurements, traffic terms, delivery timing and incident process. Compare candidates against the same criteria so a polished description does not outweigh actual results.
NextGo to the topic questions→