GUIDE EUROPE · DIGITAL SOVEREIGNTY

Network latency and European data centres: Milan, Frankfurt and Paris links

Latency is measured in milliseconds, but its effects are felt in every database query, voice call and real-time command. Distance cannot be removed, but routing and peering quality make a real difference, so the choice of city and architecture should rest on measurements rather than reputation. Here is how to plan it.

Updated october 2026
7 min read
Altovar Technical Editorial Team

1. Why latency matters beyond the data centre

Latency is the time a message takes to travel from point A to point B and back. For a static web page it is a minor detail. For a management system that runs continuous database queries, a video surveillance system, a trading platform or a plant that receives commands in real time, every millisecond adds up across every request. The result is felt as slowness, timeouts and operators waiting in front of their screens.

Many businesses assess the location of a data centre only on price or on the name of the city. The figure to look at is instead the round-trip time between the systems that talk to each other, and between users and the services they use. If this time has not been measured, the choice is still an estimate. A simple measurement, repeated at different times, changes the discussion with the supplier.

2. What determines the latency of a connection

Distance is the first factor, and it cannot be reduced to zero. Signal in optical fibre travels at about 200 kilometres per millisecond, so between two distant cities the minimum response time is fixed and determined by geography. No supplier can eliminate it. The European cities in this article are close on a continental scale, but the difference in path between one network and another matters just as much.

Beyond distance, the number of intermediate hops, the quality of links between operators and congestion at peak hours all have an effect. A network that passes from one operator to another through more intermediate exchanges can be slower than a longer but more direct path. For this reason peering, meaning the direct exchange of traffic between networks, is a relevant technical point to ask the supplier about.

3. Milan, Frankfurt and Paris: different roles

Milan, Frankfurt and Paris are among the European cities most used to host digital infrastructure. Frankfurt and Milan host Internet exchange points, that is places where many networks connect directly. This facilitates local and regional connections. Paris is another important hub, with its own network of links. In your case, what matters is knowing which cities your users come from and which services they need to reach.

There is no single best city. A business with offices and users in Italy most likely needs short links to Milan. Those serving customers in Germany or France should evaluate links to those countries. The choice should always be backed by real measurement, not by the reputation of a city. A city chosen only because it is well known may not offer the routes you expect.

4. Architecture scenarios compared

The most common architectures combine one or more sites in different ways. The table compares four scenarios on three criteria: expected latency, complexity and typical use. The values are qualitative because they depend on actual distances and the type of link available. A field measurement is always needed before deciding. Ask the supplier to explain the assumptions behind its recommendations.

The most important difference is between synchronous and asynchronous replication. With synchronous replication every write is confirmed on two sites before the result is returned: this guarantees that the data is identical, but every write pays the time of the link. With asynchronous replication the write is fast, but the second site may fall behind by a few moments. Choose according to how much data you can afford to lose and how much delay you can tolerate.

ScenarioExpected latencyComplexityTypical use
One data centre in one cityLow for nearby users, high for distant onesLowSMEs with users concentrated in one area
Two sites in the same metropolitan areaLow between the sitesMediumBusiness continuity with near real-time replication
Two sites in different cities with asynchronous replicationDepends on distance, writes stay localMedium to highDisaster recovery with data loss accepted up to a limit
Several active sites at the same timeVariable, to be managed carefully against trafficHighDistributed services with users in several countries

5. How to measure latency before choosing

Measurement should be taken from the points where users actually work, not from the supplier's server. A test from the Milan office says little about how the management system responds to someone working from a branch in another region. Repeated measurements are needed at different times of day, because congestion changes during the day. Compare the results with real working hours, not with the times when the network is least busy.

Variability also counts, known as jitter. A link with an acceptable average latency but frequent spikes can make applications unstable, especially voice calls and video. Ask the supplier to show you latency and packet loss data for the routes that matter to you, along with the measurement period. An average value without the measurement period says little about service quality.

  • Measure round-trip time from real offices and sites, across several time slots.
  • Record the variability of response time, not just the average.
  • Check the number of hops and the route with a traceroute tool.
  • Monitor packet loss during traffic peaks.
  • Test the response time of the real application, not just ping.
  • Ask the supplier for the expected network path to your main services.

6. Common mistakes in network design

The first mistake is thinking that a CDN, a content delivery network, solves everything. It may speed up static content, but a database or management system that writes data still has to reach its destination site. The second is designing replication without asking what the maximum acceptable recovery time is and how much data can be lost.

The third mistake is neglecting DNS, the system that translates names into addresses. A slow or misconfigured DNS adds time to every connection and makes it hard to see where the problem lies. The fourth is failing to document the path: when a link changes operator or city, nobody knows any longer which applications depend on it. Document the path and the connected applications, and update them at every change.

7. How to start: the first concrete step

The first step is to create a map: which sites, which users, which applications, and which data travels between which systems. For each flow, note the type of interaction, whether it is real-time or deferred, and what delay people will tolerate. This map is worth more than any comparison of price lists. Write it on a simple sheet: one row per flow is enough, with the names of the systems and the response time you expect.

Then measure response time for a week from real locations and compare it with your connectivity and data centre provider. Ask for a written proposal that sets out the architecture, the expected routes, the measures it guarantees and those it cannot guarantee. Only with this information does it make sense to choose the city, the number of sites and the type of replication.

Frequently Asked Questions

Is a closer data centre always better?

Not always. Distance between users and services matters, but so do the quality of the network and the actual path the data takes. A data centre that is further away but linked by a direct route can be faster than a nearby one with many intermediate hops. A real measurement is needed to decide.

What are DE-CIX and MIX, mentioned in the tags?

They are Internet exchange points: DE-CIX in Frankfurt and MIX in Milan. At these locations many networks connect directly to exchange traffic. Their presence can improve local routes, but on its own it does not guarantee low latency: what matters is which networks your supplier reaches there and how it routes traffic through them.

Is synchronous replication between two cities practical?

It depends on distance and requirements. Every write waits for confirmation from the second site, so the link time adds to each operation. For applications with many writes it can slow the whole system down. Synchronous replication is often chosen only between nearby sites, with asynchronous replication between distant cities.

How do I measure latency to a cloud service?

You can use measurement tools from several points, running repeated tests at different times of day. Measure both the network response time and the time of the real application, since the latter includes processing and the database. Keep the results so that you can compare periods and providers on the same basis.

How much does latency matter for a business with offices in Italy?

It matters mainly for interactive applications and for sites far from where the system is located. For an office working with a local management system it may be of little relevance. It becomes critical for voice calls, video, control systems and databases used continuously. Assess the case using measurements from your own processes.

About this content

General informational guide. Regulations and technical requirements change over time: always check the official sources and assess your specific case with a qualified technical or legal advisor.

GET IN TOUCH

Talk to our team.

We'll show how Altovar integrates into your infrastructure. No commitment, no sales pitch. Just a technical conversation.

BOOK A DEMO →