A framework for measuring Internet resilience

When One Provider Takes Down a Country

In July 2022, a Rogers outage left over 12 million Canadians without Internet or mobile service. The scale of that disruption raises a fundamental question: why do some networks bend under pressure while others snap? The answer is what we call Internet resilience — not just the ability to stay online, but to absorb shocks, adapt, and recover quickly.

Resilience isn't a single feature. It's built from many layers: the diversity of subsea cable routes, the security of BGP routing, and even the health of local markets. Think of it like psychological resilience: individual robustness matters, but true resilience only emerges from the collective, interoperable behavior of the entire ecosystem. Below, we propose a data-driven framework for making this abstract concept measurable, based entirely on public data sources.

Defining Resilience for Networks

In networking, "reliability" means things work under normal conditions, and "robustness" means things survive a traffic spike. Resilience is more dynamic — it's the ability to degrade gracefully, adapt, and recover. Our working definition is:

Internet resilience is the measurable capability of a national or regional network ecosystem to maintain diverse and secure routing paths in the face of challenges, and to rapidly restore connectivity following a disruption.

This framing ties the abstract ideal directly to metrics we can actually measure and analyze.

The Decentralized Reality

The Internet is global, but its building blocks are local. Unlike power or water grids — where a central authority can coordinate and monitor — the Internet has no global controller. It's made up of thousands of Autonomous Systems (ASes): ISPs, universities, cloud providers, enterprises. Each AS independently decides how it connects, which routes it accepts, and how it forwards traffic.

This decentralization is a strength: no single point of failure can take down the whole network. But it complicates measurement at a country level. National aggregates can mask dangerous concentration. A country may appear well-connected internationally, but if all those connections run through a handful of networks, one failure can isolate the entire nation.

The goal isn't to sever national networks from the global Internet. It's the opposite: deeper, more diverse integration makes everything stronger. When local networks invest in redundant interconnections, they improve both their own resilience and the stability of the global system.

Route Hygiene Across Layers

The Internet follows a layered model, and vulnerabilities in one layer can cascade into others. Layer-specific risks include:

• Physical layer: Geographically diverse cable routes and multiple landing stations prevent a single cut from isolating a region.
• Network layer: Multi-homing and participation in Internet Exchange Points (IXPs) give operators rerouting options when an upstream provider fails.
• Application layer: Content Delivery Networks (CDNs) and caching keep popular content reachable even when upstream routes are disrupted.
• Policy and market: Open peering policies and healthy competition foster diversity; dependence on a single ISP or cable system creates fragility.

Routing integrity itself is a matter of security hygiene. The Border Gateway Protocol (BGP) was never designed with security in mind. A single bad route announcement — accidental or malicious — can misdirect traffic or trigger outages. Two mechanisms help:

RPKI (Resource Public Key Infrastructure) lets networks publish cryptographic proof that they're authorized to announce certain IP prefixes. ROV (Route Origin Validation) checks those proofs before accepting routes. Together, they act like passports and border control for Internet routes, filtering out hijacks before they spread.

Resilience emerges when all layers work in concert. A weak point in any one layer makes the entire system more fragile. That's why our approach treats route hygiene, cable diversity, and market structure as interconnected variables — not separate concerns.

Turning network opacity into measurable signals

The most valuable resilience data — internal topologies, cable paths, peering agreements — sits with private operators who treat it as a competitive asset. Public measurement therefore depends on stitching together partial views: RouteViews and RIPE RIS expose BGP connectivity, traceroutes reveal router-level paths, and IXP or cable maps sketch the physical layer. Each source has blind spots, so resilience analysis is inherently a matter of triangulation rather than direct observation.

To make this tractable, resilience must be decomposed into concrete, quantifiable dimensions covering physical plants, interconnection patterns, topology, and routing behavior. The metrics below draw on existing research, including the ISOC Pulse framework, and stick strictly to reproducible public data sources.

Interconnection infrastructure

Networks meet in two complementary facility types: colocation centers provide space, power, and cooling, while IXPs layered on top supply the switching fabric. Resilience depends on the health of both. Relevant measures include the count and spread of IXPs normalized by population or area, the share of local networks actually peering domestically, and the mix of member types — ISPs, CDNs, cloud providers — which indicates whether content is reachable locally during international outages.

Two directional flows round out the picture:

• Local networks that peer at international IXPs, adding routing options.
• Foreign networks that peer at local IXPs, pulling content closer to users.

Balanced movement in both directions creates independent entry and exit paths. Geographic dispersal of IXPs matters just as much: a country whose exchanges cluster in one region risks a localized failure becoming a national one. Standard spatial statistics apply here, including infrastructure per capita (echoing teledensity) and per unit area. The Location Quotient (LQ) summarizes both by comparing a region's share of infrastructure against its population or area share.

In the U.S. example above, even states with the most facilities host fewer than their population share would predict — a reminder that raw counts overstate availability.

Usage-weighted and economic metrics

Spatial measures miss how infrastructure is actually used. Infrastructure Utilization Concentration applies the Herfindahl–Hirschman Index (HHI) to facility activity. For IXPs, market share can be based on peak traffic, connected ASNs, or port capacity — and the choice changes results materially. One Central European country scores 1,316 (unconcentrated) by ASN count but 1,809 (moderately concentrated) by port capacity.

Inequality measures add another lens. The Gini coefficient (0 equal, 1 fully concentrated) and its visual counterpart, the Lorenz curve, expose how unevenly infrastructure is spread across regions.

The paired charts above show that U.S. colocation concentration by geography is stark, but largely explained once population distribution is factored in.

Submarine cables and backup paths

For undersea infrastructure, resilience is the ability to survive physical damage and restore intercontinental traffic quickly. Key indicators:

• Diversity of landing stations, corridors, and operators. Single points of failure are the enemy; Tonga's total outage after one cable failure is the canonical warning.
• Fault rates and mean time to repair (MTTR). Repairs hinge on vessel availability and, often more critically, government permits, which can become the dominant delay.
• Satellite backup readiness. Though pricier and lower-capacity than fiber, LEO and MEO constellations have proven themselves in conflicts and disasters. Readiness metrics cover contracted NGSO providers, satellite bandwidth, and deployment of terminals at public and critical sites.

Inter-domain routing robustness

Above the cables and routers sits the control plane, where BGP failures can blackout traffic even when hardware is intact. Several metrics characterize this layer's fault tolerance.

AS Path Diversity counts independent routes between points. The architecture underneath matters:

• Single-homed ASes depend on one upstream — cheaper but fragile.
• Multi-homed ASes hold multiple upstreams and trade cost for redundancy.

The share of multi-homed ASes serves as a first-order resilience gauge, measurable directly from public BGP collectors. Longitudinal monitoring catches backup links that single snapshots miss. Drilling into the distribution of single-homed customers per provider surfaces systemic weak spots — the networks that would be orphaned when a particular transit fails.

For Canadian providers in July 2025, total customers are plotted against single-homed dependents. With an overall single-homing rate near 30%, the pattern recalls the 2022 Rogers outage that left over 12 million users offline.

Counting links is only half the story; the actual paths taken matter more. Path-centric metrics include:

• Path independence, assessed via Jaccard distance between AS paths. Path disjointness goes further, measuring what share of path pairs share no AS at all.
• Transit entropy, which tells whether traffic spreads evenly across providers (high Shannon entropy) or funnels through a few.
• International connectivity ratios, the proportion of domestic ASes with direct foreign links.

The polar charts above compress these same metrics for two nations with starkly different resilience profiles.

Betweenness Centrality can locate critical nodes but suffers from collector bias — ASes feeding RouteViews or RIS appear falsely central. AS Hegemony, developed by Fontugne et al., filters those viewpoints to yield a 0–1 score of true path dominance. Customer cone size from CAIDA rates an AS's downstream influence; cone rankings adapt better to resilience work when scoped to countries rather than taken globally.

Impact-weighted assessment

Treating all networks equally obscures risk. A hosting provider's failure differs from a national ISP's, so resilience metrics merit population or prefix weighting. Weighting multi-homing by users reveals how many people actually benefit from redundancy, not just how many networks claim it. Weighting by announced prefixes emphasizes players controlling more address space. The distinction separates theoretical from practical resilience: many multi-homed networks can hide a population concentrated behind one fragile single-homed ISP.

Routing hygiene and security posture

Outages frequently trace to misconfigurations or deliberate routing abuse rather than physical damage. Network hygiene — the set of practices making inter-domain routing trustworthy — is therefore a resilience property.

• Cryptographic validation (RPKI). ROA coverage tracks what share of announced address space carries valid authorizations; ROV deployment measures how many networks actually drop invalid routes, though detecting filtering in practice is hard.
• Ingress filtering (BCP-38) to block spoofed packets. Compliance testing requires active probing from within networks, performed by the CAIDA Spoofer Project. The global sample is limited, but the signal indicates both technical posture and community engagement.
• Filtering norms such as MANRS, which codifies best practices — route filtering, anti-spoofing, accurate contacts, and RPKI/IRR validation — as a public commitment rather than a technology.

Each participating network raises the collective bar. Public measurement infrastructure amplifies the effect: RIPE Atlas and CAIDA Ark probe data-plane health, RouteViews and RIPE RIS track BGP anomalies, and PeeringDB documents interconnection points. Observatories like IODA and GRIP fuse these feeds for near-real-time outage detection and incident visibility, making resilience a shared, observable property rather than a private one.

From raw routing data to actionable resilience metrics

The framework is built on publicly available data, which means the resulting metrics are reproducible and transparent. That combination matters: a resilience score that cannot be independently verified is of little use to regulators or operators who need to make decisions based on it.

The methodology translates complex routing behavior into a set of concrete, measurable indicators. These indicators cover different facets of resilience, such as the diversity of paths available, the redundancy of critical infrastructure, and the potential for a single failure to disrupt connectivity. Each metric is designed to be calculated from the same publicly available routing data sets, ensuring consistency across different regions and time periods.

By breaking resilience down into discrete, measurable components, the framework makes it possible to compare networks, regions, or even specific autonomous systems on a like-for-like basis. It also exposes the underlying structure of the Internet in a given area, revealing where the dependencies and potential breaking points actually are.

Resilience as a shared responsibility

The value of such a framework extends beyond academic interest. It gives stakeholders across the Internet ecosystem a common reference point for discussions about infrastructure strategy and risk.

• For policymakers, the metrics can guide decisions about where to invest in infrastructure or how to design policies that promote a more diverse and competitive market. Instead of relying on anecdotal evidence, they can base decisions on measurable indicators of network health.
• For network operators, the framework offers a benchmarking tool. Operators can assess their own resilience posture and compare it with that of their peers or business partners, identifying areas for improvement.
• For the broader Internet community, transparent metrics on resilience create pressure for accountability. When the strengths and vulnerabilities of a network ecosystem are visible, there is a stronger incentive to address the weaknesses.

The aim is not to assign blame or to rank networks for its own sake. It is to move the conversation about Internet resilience away from abstract concerns and toward a shared, evidence-based understanding of where the real risks lie. Only with that understanding can stakeholders collectively work to eliminate single points of failure and build a global network that is more robust and dependable.

For those seeking the full technical details, including the complete metric table and associated source code, the full paper is published as Regional Perspectives for Route Resilience in a Global Internet: Metrics, Methodology, and Pathways for Transparency, available at SSRN with the abstract ID 5376106, and was presented at TPRC23.