Observe repeated responses along a route
Mtr repeatedly probes a route and reports responses and round-trip observations for intermediate points and the destination. A router not answering a probe does not establish an identical loss rate for all forwarded traffic. Forwarding and locally generated diagnostic replies can have different priorities or rate limits. Define destination, source, protocol, IP version and observation window before interpreting the display. One alarming row is not enough to identify a defective device. Relate the observation to the application incident interval. The report is most useful for testing path hypotheses when the probe conditions and the scope of the conclusion are both explicit.
Save a report with bounded probes
The example produces a bounded ten-cycle report and avoids reverse-name lookups. Substitute a destination within the intended diagnostic scope and use only the frequency needed for the question. TCP or UDP probes and destination-port selection depend on the installed version and permissions. Default ICMP probes can encounter different filtering or routing from an application’s TCP connection to port 443. No current route or loss measurement is claimed here. Preserve command conditions, interval and whether the destination answered. Those facts prevent a later report from confusing a partial path observation with proof of a failed end-to-end application connection.
mtr --report --report-cycles 10 --no-dns example.comAvoid misreading intermediate-only loss
Suppose an intermediate row hypothetically shows forty-percent missing replies while later hops and the destination show almost none. Reply rate limiting is a possible explanation, so the row alone does not prove forty-percent forwarding loss at that router. The percentage is an interpretation example, not a benchmark. Persistent loss continuing through later observations can motivate further investigation, but consider path distribution, sample count and time variation. Marking every silent row as a broken link can incorrectly accuse equipment that forwards traffic normally. Read the destination result and the rows together as one bounded observation rather than converting each intermediate percentage into an independent definitive diagnosis.
Separate round trips from application paths
A hop’s timing generally covers the outward probe and its reply. Subtracting adjacent rows does not necessarily yield the one-way latency of the intervening link. Return paths can differ, and probes may use different equal-cost routes. Load balancing can change the visible intermediate address, so avoid forcing the report into a single fixed-path model. Mean and worst observations describe the collected sample population, not an application’s universal ninety-fifth percentile. Reverse names or geographically suggestive labels also do not prove physical placement. Preserve addresses and query conditions, then use topology information from an appropriate authority when a more specific location claim is needed.
Narrow hypotheses with independent evidence
Align the report with application connection failures, TCP retransmissions and request latency. Healthy probes to the destination do not establish healthy HTTP execution or database waiting. Conversely, limited ICMP responses can coexist with successful application traffic. Change protocol or source deliberately, one condition at a time, when testing a path closer to the application. After repair, repeat matched observations and verify destination behavior and user-visible delay. A shared report should include destination, IP version, probe method, time and sample count. Do not isolate one large intermediate number and present it as proof of device failure without the end-to-end and protocol context needed to interpret it.