With Traceroute and MTR, you can determine which nodes along the path to a hosting destination respond to diagnostic probes. MTR repeats the probes and shows whether any anomalies persist over the measurement period. A a noticeable intermediate hop However, this does not prove that the router is defective or that there is packet loss in application traffic. The diagnosis becomes more meaningful if you record the destination address, protocol, and time, and compare the responses all the way to the destination with the actual error pattern.
Define the Measurement Objective and Prerequisites
Effective trace analysis begins with specific details: From which computer is the problem occurring, which service is affected, and when does the error occur? Note the time, including the time zone, and describe the symptom—such as slow page loading or a connection being lost when accessing a website.
Hold on to that Output network fixed, such as a home connection, an office, a mobile network, or a server location. A trace from your own laptop initially answers a different question than a trace from a hosting server: Both start at different points and may take different paths to their respective destinations.
For comparison later on, you'll also need the actual destination IP address and whether it's an IPv4 or IPv6 address. Also note the application in question and, if applicable, its port—such as TCP port 443 for HTTPS. A domain name alone is not sufficient as a unique measurement identifier if it can resolve to multiple addresses.
For a website, the resolved address may belong to a CDN or reverse proxy. In that case, the measurement shows the path to that endpoint; the connection from there to the actual Origin Server This does not automatically mean it is checked. Therefore, make sure to specify whether you are checking the publicly accessible endpoint or a separately authorized Origin address.
Test only your own targets or those expressly approved for diagnostic purposes, and coordinate the scope and timing when working with production services or third-party systems. Check the installed tool version and its local help: Options and required permissions may vary depending on the system and installation. Do not grant additional permissions without verifying them just to start a measurement.
How Traceroute Reveals the Path
Traceroute sends probes with a limited lifespan into the network. For IPv4, the corresponding field is TTL, known as the "hop limit" in IPv6. Routers decrease this value as the packet travels; if it expires, the router in question can send an ICMP Time Exceeded response back to the originating system.
For the first probes, the tool sets the value low; for subsequent ones, it gradually increases it. This allows devices at greater distances to respond one after another and appear as hops. At the destination, the possible response depends on the selected probe method; a trace is therefore a representation of received responses and not a complete network map.
An asterisk initially simply means that no matching response was received for the probe in question within the waiting window. A router may filter responses or fail to send them, even though it is forwarding traffic. A lost response is also a possibility; therefore, a single asterisk does not necessarily indicate a failed device or a broken path.
The displayed Round-trip time includes the path taken by the probe to the respective responder and the return path of its response. It is not simply the round-trip time to the server. The return path may differ from the outbound path, and devices may handle diagnostic responses differently.
Therefore, if you subtract the times of two adjacent hops from each other, you won't get a reliable measurement of the delay on the single link between them. A high value at a station is a reason to examine subsequent responses and further measurements, but on its own, it is not proof of a slow router.
Distinguishing Between MTR, Traceroute, and pathping
Traceroute is useful for getting a quick overview of the visible stations along the path to a destination. However, for a transient disruption, a single run may not be sufficient: it shows only the responses to the probes sent during that run. A subsequent run may reveal a different picture.
MTR combines path tracking with repeated probes and statistically summarizes responses and response times per displayed hop. A report can thus show whether a conspicuous response pattern recurs during the observed time period. For a sporadic disruption, the correlation with the actual time of the error is particularly important.
Even a longer MTR report remains a snapshot limited to a specific time period. Its loss and time data refer to the diagnostic packets used and the responses received. It cannot be readily concluded from this that all packets from a website, database connection, or other application were handled in exactly the same way.
On Windows, it provides tracert a path indicator; pathping is an alternative for additional statistics. The latter uses its own measurement method. Therefore, do not simply compare the numbers from different tools; instead, take into account the measurement duration, objective, log, and the respective context in which the data was generated.
Probes may use different protocols depending on the tool and settings. Filters and target systems may respond differently to these protocols, so a comparison is only meaningful if it follows a documented procedure. Before taking a measurement, check the Help section of your installed version to see which options are available and what permissions are required.
Creating MTR Reports and Traces
On Linux, a good starting point for reproducible results is a MTR Report with numeric addresses. In the following examples, which are not intended to be executed, replace example.com with your own destination or one that has been explicitly approved for diagnostic purposes. First, check the locally installed version, its help documentation, and any required permissions, rather than extending permissions without verification.
The following command in the Linux terminal requests an IPv4 report with 100 measurement cycles. These options require an MTR version that supports them; they are described in the MTR 0.95 documentation.
In doing so, it selects -4 IPv4, -r report mode and -n the numerical display without name resolution for the hops. With -c 100 Specify the number of cycles. Save the complete output along with the measurement time so that you can match it to later comparisons.
One hundred cycles are a reasonable starting point, but they do not guarantee a representative failure rate. If an error occurs only rarely, multiple time-stamped reports may be more useful than interpreting a single sample. Be sure to coordinate the scope and timing, especially when dealing with third-party targets and production services: Repeated probes generate diagnostic traffic.
In Windows, you can first create a trace with numeric hop addresses in the Command Prompt. Here, too, replace the placeholder with a shared destination; this option suppresses name resolution for the routers displayed.
This trace provides a quick overview, but not the kind of repeatable statistics found in an MTR report. Along with the output, make a note of the source system and the time. This way, you can determine whether two seemingly different results were actually generated under comparable conditions.
A Detailed Comparison of IPv6 and TCP
If the service is to be accessible via both address families, test IPv4 and IPv6 separately. Record the specific destination IP address used in each case and the time of the test: A single name can point to different endpoints, and even for the same service, packets may take different network paths.
To compare IPv6 addresses, use the following sample command in the Linux terminal. You'll need a target that can be reached and an installed version of MTR with the specified options; replace the placeholder with your shared target.
Compare the report with the IPv4 results as soon as possible and correlate both with the observed application symptom. A discrepancy is an indication of an issue with that specific connection, but it is not proof of a general IPv6 error. The article on [...], describes other possible causes in a hosting context. IPv6 Issues with Servers and Networks.
If there is an HTTPS issue, a second comparison can be performed using TCP Probes to Port 443. The following Linux command—which is not executed here—requires an MTR version with TCP and port options and again uses a specified destination instead of the placeholder.
The option -T selects TCP, -P 443 the destination port. If the results of ICMP and TCP probes do not match, differences in protocol handling or filtering may be possible explanations. However, the comparison alone does not reveal which specific firewall or device is responsible.
A response to the TCP probe does not constitute a complete TLS handshake and does not prove that the website is delivering its content correctly. Therefore, you should also verify HTTPS using an appropriate application test. Conversely, a missing probe response may be caused by a rule and does not, in and of itself, indicate a failure of the entire application.
Understanding Packet Loss and Latency Correctly
Read a report covering the intermediate stops up to the destination, and look for patterns rather than a single line that stands out. Displayed Parcel loss This initially indicates a lack of responses to the respective probes. Whether this indicates a problem with application traffic must be determined through further observation.
| Observation | Plausible explanation | What Does Not Follow From This | Next step |
|---|---|---|---|
| Loss on just one stopover | The router responds to diagnostic packets with limited functionality or at a low priority. | The router inevitably loses forwarded traffic. | Check the following hops and destination, as well as the application symptom. |
| A Similar Deficit as the Finish Line Approaches | Loss or filtering may occur during transmission or when processing responses. | The first noticeable hop is certainly the source of the error. | Repeat the test shortly thereafter and compare the results with the application error. |
| No target answer | The target or a filter may not respond to the probing procedure. | The server or application has gone down. | Check the appropriate TCP port and the service separately. |
| Latency jump at a hop | There may be a delayed router response or a problem on a section of the path. | The difference reliably measures the delay of a single link. | Compare subsequent hops and the target; measure repeatedly. |
If an intermediate hop rarely responds, while later stations and the destination do respond, the router’s response processing may explain the display. However, this does not prove that every single application flow is error-free. If a similar loss pattern persists all the way to the destination, this is a stronger indication that a timely cross-check is warranted.
Even when it comes to the Latency It's the big picture that counts: If a jump disappears in subsequent hops, it may simply be that the response from the router in question is delayed. If it persists, a slower path segment could be a possible explanation. Since each time measurement includes both the outbound and return trips, the individual transition cannot be precisely measured from it.
Filtering and different return paths can also affect which responses are received. In addition, load balancing or routing changes can reveal varying hop counts. Therefore, do not blindly compare the same line number from two reports: First, check whether the destination, protocol, measurement time, and displayed path match at all.
Narrow Down Malfunctions Using Comparative Measurements
If a website is slow on only one connection, compare the affected connection to a second network as soon as possible. Use the same specific destination IP, address family, and probe type. This will minimize differences caused solely by varying destinations or measurement methods; nevertheless, be sure to note the time and the source network for both reports.
A different result from the second network may narrow down the search space, but does not prove that there is an error at the first connection. The initial networks may use different paths to the destination; temporal fluctuations or different handling of the probes are also possible. Therefore, check whether the difference occurs repeatedly and at the same time as the slow execution.
If the server is generally accessible but HTTPS is causing problems, compare ICMP probes with TCP Probes to the port actually in use. Different responses may indicate protocol-specific handling or filtering. It is not possible to determine from this alone which specific component is responsible.
Also perform a separate HTTPS application test and note whether, for example, the connection fails to establish or a page responds slowly. A TCP probe to port 443 does not simulate the complete TLS handshake or the processing of an HTTP request. Therefore, a reachable port does not mean that the website is functioning correctly.
If you experience occasional glitches, there are several things you can try Comparative measurements during and outside the error window. Record the duration and number of probes in each case: One missing response per 100 probes corresponds mathematically to one percentage point in this sample. This is neither an independent measurement nor an indication of the long-term loss rate.
Prepare a countermeasure and a support ticket
A measurement from the server to the client's IP address complements the trace from the client to the server. It examines the path originating from the server using its own probes and responses. Even though the same addresses are involved, it does not reliably reconstruct the return path of the original trace: routing and response behavior may differ between the two directions.
For a usable Countermeasure The destination address and the time of measurement must be known. Check which public client IP address is being used during the outage and whether it is permissible and appropriate to run a diagnostic test on that address. A firewall may cause the selected probes to go unanswered; in that case, the result alone does not prove that the client has failed.
If you don't have access to the server, ask the hosting provider to perform a traceback to the corresponding client IP. Specify the error message and the affected application so that support can select an appropriate test and place the result in context. If necessary, coordinate the scope and timing of the tests instead of continuously generating diagnostic traffic without being asked.
For the Support Ticket Complete, unmodified reports are more helpful than a single highlighted hop. Provide the context and describe what is actually happening in the application. The following checklist helps you classify the findings in a reproducible way:
- Source and destination IP addresses, address family, and source network or server location.
- Tool used, including its version if possible, probe log, and target port, if applicable.
- Complete reports including the date and time, time zone, and the number of cycles or measurement duration.
- Specific symptoms observed in production, frequency, and the observed error window; if available, relevant comparison reports.
Reports may contain public and internal addresses or other network-related information. Share confidential information only through an appropriate support channel, and, if necessary, clearly mark any information that has been intentionally omitted. In contrast, an isolated loss figure without a target, time, or log can hardly be meaningfully correlated with an incident.
Next Steps After the Network Check
If recurring anomalies up to the destination coincide with the application error, it is worth conducting an in-depth investigation of the network path and possible filters. To do this, compare logs and measurement directions, and forward the findings to the responsible service providers. The first anomalous hop alone does not indicate a faulty router or definitively identify the responsible provider.
A unobtrusive report does not rule out a hosting issue: It merely describes the responses to the probes used during the measurement period. Next, depending on the symptom, check the name resolution, the TLS connection establishment, and the HTTP response. Be sure to record the destination address and the time so that the results can be compared with the network report.
If HTTPS returns a response but certain functions remain slow or fail, also investigate the application and its dependencies, such as database access. When accessing the server, you can bpftrace for server-side diagnostics provide additional insights into processes and kernel events. This is not a substitute for measuring the path between the client and the server.
When dealing with size-related transmission issues, also consider the MTU: Simple standard tests are not sufficient to rule out such errors in general. If larger packets or filtered error messages could be a factor, treat in-depth MTU tests as a separate diagnostic step and coordinate them with the relevant service providers.
Don't change multiple components at once just because a report looks suspicious. First, document your hypothesis, perform an appropriate individual test, and compare the results with the original error pattern. This way, you can determine whether a corrective action has affected the observed problem or has merely altered a single diagnostic response.
Sources and Current State of Knowledge
Status of the research:
Source base and tool options are based on the research conducted on November 13, 2026; MTR examples refer to the documented options in version 0.95. Check the installed version and local help before use.
https://github.com/traviscross/mtr/blob/v0.95/man/mtr.8.in
https://www.cloudflare.com/en-ca/learning/network-layer/what-is-mtr/
https://developers.cloudflare.com/speed/troubleshooting/slow-website/
https://man7.org/linux/man-pages/man8/traceroute.8.html
https://www.rfc-editor.org/info/rfc792/
https://learn.microsoft.com/en-us/windows-server/administration/windows-commands/tracert
https://learn.microsoft.com/en-us/troubleshoot/windows-client/networking/diagnose-packet-loss




