Summary

  • LACNIC’s study reports at least 138 potential routing-loop cases, but its intra-AS count moves from 24 to 83 when the attribution rule changes. The same 138 candidates are being classified differently.
  • Public Atlas records show individual repeated hop sequences. They do not establish the study’s full sample, duration, service impact or the legal identity of the network responsible.

Analysis

The route comes before the owner

In measurement 215329771, one ICMP traceroute from RIPE Atlas probe 60205, recorded on 24 September 2026, alternates between 45.186.15.34 and 45.65.244.151 at every displayed TTL from 9 through 30. A second result, measurement 215330052, repeats a three-address sequence over the same range, with one probe timing out at one hop. Those are stronger examples than a single address repeated down a trace: a periodic sequence is visible in the returned path itself.

LACNIC’s 8 October article describes a broader study. It says 1,476 IPv4 prefixes announced by 197 Dominican Republic ASes were examined. At least 40 addresses in each prefix received ICMP probes. Where a probe returned Time Exceeded, the address was retained and another address in that prefix was selected at random. The authors then obtained 234 destinations for traceroute measurements, using two RIPE Atlas probes, ICMP and TCP methods, and later custom checks.

The published counts need to be read as the output of that method, not as a census of damaged networks. LACNIC reports at least 138 cases with forwarding patterns consistent with potential loops. It first classifies 24 as intra-AS and 14 as inter-AS; 100 remain unresolved because private addresses or timeouts prevent reliable identification. It then applies a different rule: if the last public hop before a private-address section and the destination prefix belong to one ASN, classify the case as intra-AS. The result becomes 83 intra-AS and 41 unclassified.

Both versions total 138. The second rule does not discover 59 new loops. It assigns 59 of the earlier uncertain cases to a category by carrying an ASN label across a private-address segment. That may be a useful hypothesis for troubleshooting. It is a different evidentiary statement from seeing every relevant hop inside a publicly attributable AS.

A visible sequence is not a complete topology

The Internet Protocol’s TTL limits how far a packet can be forwarded. RFC 1812 requires forwarding routers to reduce TTL, and identifies TTL as a mechanism for terminating routing loops. Traceroute turns the resulting responses into a hop-by-hop view. The view is valuable because it shows what answered a probe at each limit. It is still a view assembled from responses, not a map of every forwarding decision.

The difference matters in measurement 215333649. Its single public result alternates between 172.17.123.66 and 172.17.117.105 from TTL 9 through 30. This is a clear repeated pattern, but both addresses are private. Private addressing does not mean “outside the network” or “unimportant”; operators routinely use it internally. It does mean that the addresses alone do not provide a public ASN mapping. The destination prefix or surrounding public hop may suggest an administrative boundary, but a suggestion must be labelled as an inference.

Measurement 215326566 offers the opposite caution. Every TTL from 9 through 30 returns 38.131.62.230 in one Atlas result. That repeated responder is notable. It does not display the same alternating multi-router cycle, and LACNIC lists the example as unclassified by loop topology. Treating every repeated address as a confirmed loop would erase a distinction the study itself preserves.

A route observation, an AS assignment, a forwarding failure and a customer incident therefore need separate evidence. One can have a repeated hop pattern without knowing the router’s administrative owner. One can infer an AS association without knowing whether users lost traffic. One can establish a loop candidate without establishing its cause, duration or consequence.

Publish the classification rule beside the count

The study’s most consequential result may be methodological: the owner-level story changes when the classifier changes. A reader should not have to infer that from two paragraphs. A durable result record should include the candidate identifier, destination prefix, source probe, transport, UTC timestamp, each observed hop and timeout, the cycle-detection rule, the AS-mapping snapshot, and the confidence assigned to the attribution. If a private segment is bridged to an ASN through a neighboring public hop, the record should say so explicitly and keep that conclusion separate from the raw sequence.

The page says the tests were repeated at different times and that observed patterns can persist. The public prose does not give a complete run inventory, the observation window, or a per-destination matrix across probes and transports. That is not a reason to dismiss the study. It is a reason to distinguish its reported finding from what an outside reader can independently reproduce from the listed examples. A single Atlas result is a timestamped trace, not a time series.

The same distinction applies to the two vantage points. Probe 60205 was connected through AS6400, Claro Dominicana; probe 32174 was connected through AS267705. That design can reveal whether a pattern depends on where the measurement begins. It does not establish that every destination was measured successfully from both probes with both protocols. The public summary does not provide that matrix, so no broader coverage claim follows from the design description alone.

Sources