Summary

  • RHnet’s January and August 2002 incident records show that its nominal Síminn backup also relied on CANTAT-3. The two logical paths therefore shared a physical failure point.
  • When the cable system failed in August, satellite service became the emergency alternative, but the log records a roughly 600 ms round-trip time and a separate delay before RHnet itself reconnected.

A backup can disappear with the primary

The revealing detail in RHnet’s log is not simply that CANTAT-3 failed. It is the explanation for why a second provider did not preserve the research network’s international connection. The January entry says RHnet’s external link was cut at about 23:26 and that the reserve route through Síminn to the United States was also broken; both appeared to depend on CANTAT-3. The log says other Icelandic Internet services also seemed unavailable at that moment, but it gives neither a national measurement nor a measured duration.

On 28 August, the dependency became more explicit. A CANTAT-3 fault at 06:13 left RHnet without a connection to NORDUnet because Síminn’s backup also used the cable. The word “backup” described a different service path, not an independent undersea route. It was redundancy at one layer and a shared risk at another.

That distinction is easy to lose in network diagrams. A path may have a separate carrier, router, contract or point of handoff while still converging on the same cable, landing station, power feed or terrestrial segment. The RHnet record does not map every part of the architecture. It does establish the critical shared dependency that mattered during this incident: the alternate service RHnet expected to use could not bypass CANTAT-3.

Satellite restored reach, with a different service profile

The operational sequence also resists a neat “cable failed, satellite took over” summary. Síminn had satellite service working by 15:30 on 28 August, but RHnet remained disconnected because it was dealing with router problems. At 17:20 RHnet connected over satellite. The log reports an approximate round-trip time of 600 milliseconds. That is one observation in one incident, not a general latency measurement for Iceland or a full account of application performance.

RHnet moved back to CANTAT-3 at 22:23 after the route was restored. The log then records planned disconnection and satellite fallback during repair work in early September, followed by restoration of the direct route to NORDUnet on 9 September. It shows operational recovery as a sequence of carrier readiness, RHnet equipment configuration and cable availability—not as an automatic property of having purchased a backup.

A notice copied into the log on 29 August described damage between the Faroe Islands and Britain. The reported fault involved insulation and leakage current; the optical fibres were said to remain intact, with power supplied from both ends. Service quality was reduced pending repair. This matters because “cable failure” can refer to a power or protection fault that interrupts service without a snapped optical fibre. The notice does not establish that every traffic path failed in the same way or for the same duration.

A second cable changed the planning problem

CANTAT-3 had carried Icelandic international traffic since 1994. Later, the EFTA Surveillance Authority’s 2006 decision documented a history of failures and described satellite as an imperfect second connection: its cost was expected to rise, and its delays made it unsuitable for some Internet traffic. The decision also records that Farice had been established in 2002 to develop another undersea route connecting Iceland with the Faroe Islands and Britain, with public guarantees and support forming part of the project’s financing.

The Icelandic Government announced FARICE-1’s landfall in September 2003 and expected service in January 2004. Its announcement described capacity more than one hundred times that of CANTAT-3 and framed the new system as a gain in international capacity and security. Landfall was a construction milestone, not proof that service began on that date. Nor does a capacity figure tell us how much bandwidth was used or how reliably it performed.

In 2006, the Government said that after FARICE-1 entered service, the satellite fallback then available could barely carry more than telephone service, and it announced planning for an additional cable. This is evidence of how officials described the resilience gap at that time, not a universal service test. Multiple cables create options only when their routes, landing facilities, power, backhaul and operating arrangements avoid the same failure domains.

What the incident can—and cannot—prove

The RHnet log is unusually valuable because it captures the mechanics of recovery: a shared physical route, a satellite alternative, router trouble and a measured but high round-trip time. Its scale is also narrow. RHnet was a research and education network; this record cannot stand in for every Icelandic operator, customer or service. The January note’s reference to other services is an observation, not a countrywide outage dataset.

The broader lesson is operational. Redundancy needs to be tested against the failure that can remove both paths, and the fallback must be evaluated by the service it can actually carry. A contract naming a second provider does not prove physical diversity. A second cable’s advertised capacity does not prove independent backhaul, available capacity or continuity under repair. The useful evidence is the route beneath the diagram and the behavior seen when it is needed.

Sources