PATTERN Cited by 1 source
Redundant ISP absorption of partial outage¶
Pattern: In a region with multiple independent ISPs, a catastrophic failure at one ISP does not fully take the region offline — traffic from the other ISPs continues, producing a region-level outage that is proportional to the failed ISP's share of local traffic, not total.
Where many region-scale outages look like a cliff-edge collapse to zero, this pattern produces the characteristic staircase shape: the ISP with the failure goes near-zero, and the region's aggregate drops by the proportional share of that ISP — leaving survivors to carry what traffic they can.
Canonical instance: U.S. Virgin Islands, March 24, 2026¶
On March 24, 2026, the U.S. Virgin Islands experienced a significant power event: a loss of generation at the Richmond Power Plant combined with damage to an underground cable. The impact on Internet connectivity was asymmetric by ISP and by island — exactly the shape this pattern describes:
- VI Powernet (AS14434), the primary ISP for the U.S. Virgin Islands, dropped to near zero starting ~12:15 local time (16:15 UTC).
- St. Thomas saw traffic fall by ~60% — not 100%, because other ISPs continued serving the remaining ~40% of traffic.
- St. Croix saw traffic fall by only ~40% — even less of a drop, because the share of St. Croix users on VI Powernet is lower than on St. Thomas.
The Cloudflare Radar team's explicit framing: "Although VI Powernet experienced a near-complete outage, traffic from St. Thomas only fell by around 60%, and approximately 40% from St. Croix due to the presence of other providers."
Why the pattern matters¶
Two adjacent truths are captured here:
- Redundancy at the ISP layer is real. When multiple independent ISPs serve the same geography on independent transit, last-mile, and power dependencies, a failure at one does not propagate to all.
- Redundancy is asymmetric. The degree of protection the region gets depends on the share of traffic carried by each ISP — a region where one ISP holds 90% market share gets much less protection than one where the share is evenly split. St. Thomas (higher VI Powernet share) saw a larger drop than St. Croix (lower share).
For capacity-planning and reliability-engineering purposes, this is the same shape as homogeneity penalty on large distributed systems: concentrating load on one dependency increases the impact radius when that dependency fails.
Pre-conditions for the pattern¶
For a region to get the benefit:
- Multiple independent ISPs operating in the geography.
- Independent upstream transit — if ISP A and ISP B both peer to the same tier-1 transit that also failed, they correlate and the pattern breaks. In the USVI case, VI Powernet's failure was tied to local generation + cable damage specific to that ISP's infrastructure; competing ISPs presumably had independent last-mile + power arrangements.
- Independent physical infrastructure at the last mile — two ISPs sharing the same aerial-fibre route or substation are not independent for storm / grid / cable-damage events.
Architectural implications¶
For users / enterprises in the region¶
- Multi-homed connectivity via two independent ISPs is a load-bearing reliability strategy. The cost is double-ISP subscription + upstream routing complexity (BGP multi-homing for enterprise, or SD-WAN / cellular-fallback for SMB and home).
- Customer's own infrastructure fails open only if the user is multi-homed. For single-ISP customers, a 60% regional drop could still look like 100% for them.
For SaaS providers / CDNs serving the region¶
- Per-AS telemetry matters — a region-aggregate drop of 60% hides the 100% drop on one AS and 0% drop on another. Alerting on region-level metrics alone misses the per-AS detail.
- Customer impact attribution — when users complain, knowing the customer's AS is key to attributing their problem to the ISP-level event rather than to the SaaS itself.
For regions with one dominant ISP¶
- The pattern breaks down. A near-complete failure at the dominant ISP looks like a region-level cliff-edge collapse (this is the shape in many national-carrier-failure events).
- Encouraging ISP diversity is a public-policy lever for Internet resilience.
Composes with¶
- patterns/multi-region-failover — the upstream analogue of ISP redundancy at the cloud-region level.
- concepts/correlated-failure — the failure-mode the ISP redundancy partially prevents.
- concepts/blast-radius — per-ISP outage blast radius is narrower than region blast radius when ISPs are independent.
Seen in¶
- sources/2026-04-28-cloudflare-q1-2026-internet-disruption-summary — canonical wiki instance. On March 24, 2026, the U.S. Virgin Islands experienced a near-total VI Powernet (AS14434) outage following generation loss at the Richmond Power Plant plus underground cable damage. Aggregate island-level impact: ~60% traffic drop on St. Thomas and ~40% on St. Croix — proportional to each island's share of users on VI Powernet, with other providers absorbing the residual traffic. The pattern is also implicit in cases where no single-ISP failure caused region-scale total blackout (even WACS-induced Republic of Congo traffic was cushioned from 100% loss by Congo Telecom's activation of backup solutions, landing at 82% below expected — not 100%).