Hi Members,
I am investigating a network performance issue where latency measurements collected through RIPE Atlas show significant fluctuations during certain periods of the day, even though the monitored destinations remain reachable and the overall network connection appears stable. I have been using RIPE Atlas measurements to observe connectivity between several network locations and compare the results over time, but the latency patterns are becoming difficult to interpret. During quieter periods, the round-trip times are relatively consistent, with only minor variations between consecutive measurements. However, when network activity increases, the results sometimes show sudden latency spikes that last for several measurement intervals before returning to their previous levels. These changes do not always coincide with packet loss or a complete connectivity failure, which makes it difficult to determine whether the issue reflects genuine network congestion, temporary routing behavior, or another factor affecting the measurement results.
The main difficulty is that the performance degradation does not appear consistently across every measurement target. Some destinations continue to respond with relatively stable latency while others experience substantial increases during approximately the same period. I initially suspected that the issue might be related to the local network connection or the performance of the system running my monitoring tools, but the differences between destinations make that explanation less convincing on its own. I want to understand whether these variations could indicate congestion on a shared network segment, differences in the paths selected by upstream networks, or a problem occurring closer to a particular destination. Since RIPE Atlas provides measurements from distributed probes, I believe comparing results from multiple probes could help narrow down the affected part of the path. However, I am not entirely certain how to distinguish a meaningful network performance problem from normal variation between probes that use different providers, routing policies, and connection types.
I have also noticed that interpreting the latency graphs can be challenging when individual measurements show sharp increases but the surrounding results remain normal. For example, a series of measurements may show a relatively stable baseline, followed by several unusually high round-trip times, and then an almost immediate return to the previous latency range. In some cases, the elevated values appear to affect multiple consecutive measurements, while in others the change is brief enough that it could be overlooked when reviewing a longer time range. I would like to determine whether these patterns are sufficient to suggest temporary congestion or whether they could result from factors such as ICMP response prioritization, probe scheduling, or the way intermediate routers handle measurement traffic. I understand that a traceroute or ping measurement does not necessarily represent the performance experienced by every application, but I am looking for a reliable method to evaluate these observations without drawing conclusions from isolated spikes that may not reflect the actual forwarding performance of the network.
Another part of the investigation involves comparing measurements collected at different times and from probes located in different networks. The latency increases seem more noticeable during periods of higher general network activity, but I have not established a direct relationship between those periods and a specific link or autonomous system. I am considering whether the most useful next step would be to run coordinated measurements toward the same destination from several probes and compare their results over identical time windows. I would also like to understand how much confidence I can place in a comparison when the probes have different access technologies, upstream providers, or geographic locations. If several probes show a similar increase at approximately the same time, would that provide meaningful evidence of a shared upstream issue, or could the results still be explained by independent routing changes? I want to avoid treating a correlation between measurement results as proof of a common cause, especially when the available data does not directly reveal utilization or queueing conditions on the intermediate links.
I am hoping that experienced RIPE Atlas users can recommend a practical approach for analyzing this type of intermittent latency variation. In particular, I would appreciate advice on selecting suitable probe sets, choosing an appropriate measurement frequency, and deciding whether ping measurements alone are sufficient or should be supplemented with traceroute measurements. I am also interested in knowing which aspects of the results are most useful when trying to identify where latency begins to increase along a network path, especially when intermediate hops do not respond consistently or display higher response times than the final destination. If there are established practices for comparing round-trip-time distributions, identifying recurring time-based patterns, or separating measurement noise from sustained performance degradation, those would be helpful as well. My intention is to build a more dependable picture of the network behavior over time rather than increasing the number of measurements without a clear analysis strategy.
Has anyone encountered a similar situation where RIPE Atlas measurements reveal recurring latency spikes but connectivity remains available and the affected destinations do not all behave the same way? I would appreciate suggestions on how to structure the investigation so that the results can help identify whether the problem is associated with a particular network path, a shared upstream segment, or normal differences in how probes and routers handle measurement traffic. I can provide anonymized measurement IDs, destination information, timestamps, and relevant latency graphs if those details would help others assess the pattern. I am particularly interested in learning which evidence would be strong enough to justify investigating a specific network segment rather than simply observing the fluctuations. Any recommendations for a systematic troubleshooting process would be valuable, because I want to use RIPE Atlas data to understand the underlying performance behavior accurately instead of relying on assumptions based on a few unusually high latency readings. Sorry for long post!