ARCOS

Based on 41 RIPE Atlas measurements from GeoCables monitoring infrastructure, March–April 2026.

The Americas Region Caribbean Ring System — ARCOS-1 — has been in commercial service since September 2001. Twenty-four years. By the standards of the submarine cable industry, where a planned design life of 25 years is the norm and many cables are retired well before that mark, ARCOS-1 is at the late end of its first useful life. It connects sixteen countries around the Caribbean basin: the United States (with two landings, Florida and Puerto Rico), the Bahamas (three islands), the Turks and Caicos, the Dominican Republic, Curaçao, Venezuela, Colombia, Panama, Costa Rica, Nicaragua, Honduras, Guatemala, Belize, and Mexico. The cable spans 8,704 km and was originally laid as a closed ring — a topology that lets traffic flow in either direction around the basin and survive a single fibre cut anywhere along the path.

Across 41 measurements between Puerto Rico (Isla Verde) and Guatemala (Puerto Barrios) — the two cities at opposite ends of our probe coverage on this cable — the minimum round-trip we observe is 52.22 ms. The theoretical physics floor for ARCOS-1's full 8,704 km is 85.19 ms. Our measurement sits at 0.613× that floor. Below the floor. Roughly 39% faster than the cable's geometry would allow if traffic actually traversed the full ring.

Below the floor means a chord, not a loop

The number doesn't violate physics. It tells us something about routing. When a packet leaves San Juan for Guatemala, it does not travel three-quarters of the way around the Caribbean basin to reach a country that sits more or less directly opposite. It takes the chord — the shortest segment of the ring that connects the two endpoints — and arrives in roughly the time the geometry of that chord allows. Puerto Rico to Guatemala measured along a great-circle path is approximately 2,400 km of physical distance. The optical floor for that geometry, accounting for the ~1.47 refractive index of single-mode fibre and a typical 10–15% slack from non-straight cable routing, lands somewhere in the 50–55 ms range. Our 52.22 ms minimum is consistent with traffic using the western half of the ARCOS ring directly — likely the segment Puerto Rico → Dominican Republic → Turks & Caicos → Bahamas → Florida → Mexico → Belize → Guatemala, or the equally plausible southern arc through Curaçao and the Central American coast.

This is what the "sub-floor" pattern almost always reveals on a ring cable: the listed cable length is the engineering total, not the operational shortest path. Traffic prefers the chord. The full-loop geometry only becomes relevant in failure modes, when a cut on the preferred chord forces packets onto the longer alternate arc. We see the same arithmetic on other ring systems we've published — FLAG North Asia Loop, the 2001-vintage Pacific ring, shows similar below-floor measurements on its East Asia chord segments. ACE's 19-landing architecture produces below-floor numbers along the West African coast for the same reason.

Two directions of the same path, two different worlds

The most striking thing in our ARCOS-1 dataset is what happens when we split the measurements by direction. Between the same two cities — Isla Verde and Puerto Barrios — the forward and reverse measurements look like they came from different cables.

The forward direction — westbound from Puerto Rico into Central America — has a standard deviation of 11.9 ms across our measurements. The minimum is 52.22 ms but the maximum touches 102 ms, indicating that some of the time the packet is rerouted onto a longer path, possibly via the southern arc of ARCOS or onto a different cable system entirely. Forward routing is BGP-volatile: the operator at Isla Verde is selecting between multiple announced paths to Guatemala based on whatever best-path algorithm and policy preferences are currently in effect, and that selection appears to flip among at least two distinct route options.

The reverse direction — eastbound from Puerto Barrios back to Puerto Rico — is essentially locked. A standard deviation of 0.18 ms across twenty measurements means every single packet took the same path, every single time, with sub-millisecond clock-and-jitter noise as the only source of variance. This is among the lowest standard deviations we have ever measured on any submarine cable in any region. It suggests the Guatemalan upstream — likely a single peering arrangement at Puerto Barrios — has exactly one announced path back to Puerto Rico, or has set hard local preferences that prevent any rerouting.

This kind of asymmetric stability is unusual. On healthy modern cables we typically see both directions converge to similar standard deviations after BGP settles, sometimes within a few hundred microseconds of each other. The 11.9 ms / 0.18 ms split on ARCOS-1 is a fingerprint of the commercial structure of the Caribbean's internet: traffic westbound from PR has multiple commercial alternatives (US peering, alternative cables, multi-hop carrier paths), while traffic eastbound from Guatemala has effectively one. The cable's age is one factor, but the deeper cause is the asymmetric carrier ecosystem at each end.

Sixteen countries, one non-common-carrier ring

ARCOS-1 was laid in two phases. Phase 1 covered the western and northern arc — the Florida-to-Mexico segment with the major Bahamas and Caribbean island landings — and entered service in September 2001. Phase 2 closed the ring through the Central and South American coast, completing in March 2002. The result was the most landings-per-km of any cable in service at the time: twenty-four landing stations along 8,704 km, an average of one landing every 360 km of cable.

Architecturally, this is a different model from the long-haul trunks we've covered elsewhere — AMX-1 (2014, Latin American coastal, eight-country) is the closest comparable in region but uses a more linear path, and modern hyperscaler-funded systems like Firmina tend to deliberately minimise landings to reduce operational complexity. ARCOS-1 went the opposite way: maximum reach, every Caribbean basin country touched, ring topology so any single cut can be routed around. The cable was built and is operated on a non-common-carrier basis, which in practice means that capacity on ARCOS is sold and held through commercial agreements with the consortium of carriers that funded it, rather than via open wholesale tariffs.

For perspective on lifespan: PC-1, the 1999-vintage Pacific ring connecting Japan and the US, is at 26 years of service and showing the standard signs of an end-of-life cable — narrowing operator preference, gradually thinning capacity allocations, and increasing variance in measurements as routing diversifies away from it. APCN-2 (2001, Asian ring) is at 25 years and largely demoted to backup role for most of its operators. FLAG North Asia Loop, also 2001, is in the same cohort. ARCOS-1 sits in this same generational bracket and is approaching the same crossroads — replacement systems exist (Caribbean operators have signed onto newer cables in the past five years), but the ring's reach into smaller economies that newer cables don't bother serving keeps it relevant.

What we watch on a 24-year-old cable

Two specific things. First, whether the forward-direction standard deviation on Isla Verde → Puerto Barrios stabilises or grows. A widening sd on a single direction is the signature of a cable being progressively de-preferred by upstream operators — they shift more and more traffic onto younger alternatives, and the remaining traffic on the older cable shows up across an increasing diversity of paths. Second, whether new measurements continue to come in below the floor. Sub-floor patterns survive only as long as carriers continue to find a chord-of-the-ring shorter than the alternative routes. Once a clearly preferred bypass cable is in service, the chord stops being the optimal path and ARCOS measurements would start sitting at or above the floor — a quiet signal that the cable's operational role has transitioned from primary to secondary.

Both signals are worth watching not as predictions of failure but as the natural lifecycle markers of a long-running ring system. ARCOS-1 has carried Caribbean internet traffic for almost a quarter-century. Its measurements today still pass through fibre laid in the months after September 11, 2001, in a Caribbean basin whose internet economy did not yet exist as we recognise it. That continuity is, in its own way, the story.