Internet QoS measurement can inform decisions about network performance, service quality and user experience—but a result is only useful when its method and limitations are clear. For industrial automation engineers, system integrators and operations leaders assessing connectivity, transparent measurement practices make it easier to distinguish network conditions from test design, endpoint constraints or device effects.
A 2020 code of conduct developed by France’s communications regulator, Arcep, sets out practices for tools that measure fixed and mobile internet quality of service (QoS) or quality of experience (QoE). Its central principle is straightforward: publish enough information for others to understand how tests were conducted, evaluate the reliability of results and identify factors that may affect comparisons.
Why transparency matters in internet QoS measurement
A reported download speed, latency figure or streaming score is not self-explanatory. Results can vary with the protocol, test duration, number of connections, selected server, IP version and software or hardware running the test. If these details are hidden, it becomes difficult to tell whether two results are genuinely comparable.
The code applies to web-based and installable tools, Android and iOS applications, and hardware probes. It covers both the measurement protocols and the later publication of aggregated findings. Tools declaring compliance are expected to document their methods, update published methodology and test-server information every six months, and provide additional details when requested by Arcep.
Make test protocols understandable and repeatable
For download and upload tests, documentation should describe the protocol and ports used, the number of simultaneous TCP connections, test duration and any data-volume limit. It should also disclose whether traffic is encrypted, whether slow start is excluded, which IP version is used and how that version is selected. These details matter because they shape what a test measures and how its displayed throughput should be interpreted.
Tools should explain their indicators rather than presenting numbers without context. For example, a result might represent average throughput across a test or throughput over its fastest portion. Those are different summaries of performance. The code also sets minimum robustness criteria: for speed tests, the required duration or exchanged data volume depends on the tool type. It calls for a single-thread option in relevant applications, even where multi-thread testing is the default.
Latency reporting needs similar clarity. A methodology should disclose the protocol, ports, number and size of test packets, timeouts, encryption and IP version. The code specifies that latency should not be measured using ICMP, and that at least 10 unit tests should be conducted. The median latency should be shown when the test finishes. Explaining how latency, jitter or packet loss indicators are calculated helps users interpret the output correctly.
Web browsing and streaming tests need context
For web-browsing tests, tools should identify the websites tested, explain timeouts and describe cache handling. The code says not to use operators’ own portals as test sites and sets a timeout of less than 30 seconds for each website. Published indicators should explain what is included in page-load time—for instance, whether advertisements are excluded.
Streaming assessments should identify the platforms and, where applicable, resolutions tested; explain how videos are selected; and state test lengths and timeouts. The traffic should use the encryption applied by default on the platform being tested. For every test category, clear definitions help readers understand whether a result reflects a particular test scenario or a broader experience.
Check test-server capacity and selection
A measurement can be limited by the test server rather than by the connection under examination. The code therefore calls for tools to disclose information such as server host, location, connection capacity, port, supported IP versions and how the default server is selected. If server choices vary between tests, the relevant details should be available for each test.
For web, installable and mobile application tools, the code sets a minimum internet connection capacity of 1 Gbit/s for test servers and requires at least 20% of servers to support IPv6. Server information should be documented in a published annex. These requirements make it easier to identify a potential bottleneck and understand whether server selection could influence a result.
Interpret aggregate QoS results carefully
Publishing individual measurements is not the same as producing a reliable comparison. Before creating aggregate results, tools need data-processing methods that remove false, manipulated or irrelevant measurements. In particular, measurements should be excluded when the target server is a limiting factor, such as when its capacity is below or equal to the line being tested.
Publications should state the period covered and the number of tests behind each reported figure. They should also describe data adjustments and disclose potential sources of bias. For mobile results, relevant context can include test location, population density, whether tests took place while moving, operating system, device mix and the share of tests conducted over IPv4 or IPv6. Fixed-network comparisons should flag when combining technologies introduces significant bias.
These disclosures do not remove every limitation. They give readers the information needed to assess representativeness and comparability rather than treating an aggregate score as a complete description of service quality.
Practical implications for automation teams
Industrial sites may rely on internet connectivity for remote access, cloud services or other operational needs, but the code does not prescribe an industrial network design or guarantee performance for a particular application. Its value is methodological: it provides a framework for asking better questions when reviewing external connectivity measurements.
- Review the test method: Check protocols, duration, connection count, IP version and indicator definitions.
- Assess the endpoint: Look for server capacity, location and selection details that could affect results.
- Check the sample: Confirm the reporting period, test counts and processing approach behind aggregates.
- Read the caveats: Consider device, operating-system, location and technology differences before comparing results.
Reliable internet QoS measurement depends on transparent methods, adequately capable test servers and careful treatment of collected data. For technical decision-makers, these practices make performance reports more interpretable and comparisons more defensible. The practical takeaway is to evaluate not only the reported number, but also how it was produced, what it represents and which limitations may shape it.
Access Official Technical PDF