Routing Transients in Internet Networks

• Routing transients are phenomena where Internet segments exhibit partial connectivity due to routing dynamics, policy variations, and operational events, manifested as peninsulas and islands.
• Detection algorithms like Taitao and Chiloe analyze bidirectional reachability across multiple vantage points to efficiently identify and classify these transient events.
• Empirical studies show that most peninsulas last 20–60 minutes, with persistent events disproportionately affecting network monitoring and informing critical policy debates.

Routing transients are persistent or temporary phenomena in the Internet core where some networks are partially reachable from only a subset of the Internet, disrupting the expectation of uniform, global connectivity. Unlike full outages, routing transients yield partial connectivity—either limited by topology, policy, architectural constraints, or operational events—manifesting as peninsulas (partial reachability) and islands (complete isolation from the core except self-reachability). Contemporary studies characterize and systematically measure these transients, revealing their high prevalence and profound operational and policy implications (Baltra et al., 17 Jan 2026, Baltra et al., 2024, Baltra et al., 2024).

1. Formal Foundations of Partial Reachability

Recent research establishes the Internet core as the unique strongly connected component (SCC) of active public addresses exceeding 50% of globally routable space, based solely on bidirectional reachability. The set-theoretic backbone comprises:

• Active IP Addresses ($V$): Public addresses responsive to active probes over days or weeks.
• Bidirectional Routability: Two nodes are mutually routable if each can initiate and receive traffic from the other.
• Internet Core: The SCC $C \subseteq V$ such that $|C| > 0.5 \, |V|$. This "majority SCC" criterion ensures uniqueness; no two disjoint sets can both constitute a core.
• Peninsula: A block $b$ is a peninsula at measurement round $i$ if there exists a nonempty, proper subset of vantage points $O_{i, b}^{up}$ (those able to reach $b$) such that $0 < |O_{i, b}^{up}| < |O_{i, b}|$.
• Island: A vantage point $o$ at round $i$ is an island if it can reach fewer than half the core's blocks $|B_{i, o}^{up}| < |B_{i, o}|/2$. A strict case is an "address island," where $|B_{i, o}^{up}| = 0$, indicating complete isolation.

This approach is purely topological—free of registry, allocation, authority, or national hierarchies (Baltra et al., 17 Jan 2026, Baltra et al., 2024).

2. Routing Transients: Definitions and Distinctions

Routing transients are formally associated with short-lived peninsulas. Specifically, transients correspond to events where partial connectivity arises due to routing dynamics, such as BGP convergence delays, ephemeral policy changes, or temporary disruptions, primarily lasting 20–60 minutes (2–6 typical measurement rounds). Persistent peninsulas (events lasting ≥5 hours) are generally attributed to structured policy decisions, peering disputes, or long-term misconfigurations (Baltra et al., 2024, Baltra et al., 17 Jan 2026).

Unlike full outages (where $|O_{i, b}^{up}| = 0$ for all $b$ and $i$), routing transients occur when some vantage points observe reachability while others report loss—reflecting incomplete propagation, filtering, or asymmetric failures.

3. Algorithmic Detection: Taitao and Chiloe

The quantification of routing transients employs two primary algorithms:

• Taitao (Peninsula Detector):
  • Iteratively, for each round $i$ and block $b$:
  • Identify the set of valid observing VPs $O_{i, b}$.
  • Mark $(b, i)$ as a peninsula event if $0 < |O_{i, b}^{up}| < |O_{i, b}|$.
  • Discard one-round events; retain events lasting ≥5 hours for high-confidence cross-validation.
• Chiloe (Island Detector):
  • For each observer $o$ and round $i$:
  • Count reachable core blocks $|B_{i, o}^{up}|$.
  • Mark $(o, i)$ as an island if $|B_{i, o}^{up}| < |B_{i, o}|/2$. For address islands, enforce $|B_{i, o}^{up}|\to 0$.
  • Evaluate event persistence by enforcing minimum duration thresholds.

Both algorithms operate efficiently in streaming mode and are deployable over large-scale, multi-VP measurement datasets (Baltra et al., 2024).

4. Empirical Measurement and Validation

Key platforms for routing transient detection include:

Platform	Vantage Points	Target Set	Frequency	Measurement Type
Trinocular	6 datacenter VPs	5M IPv4 /24 prefixes	11 min rounds	ICMP echo probes
RIPE Atlas	~12,000 VPs	13 DNS-root IDs	5 min rounds	DNS queries
CAIDA Ark	171 global VPs	All routed /24s	daily	Traceroute

Validation leverages cross-platform consistency checks:

• Taitao vs. CAIDA Ark: Recall ≈ 0.94, precision between 0.42 and 0.82, F₁ ≈ 0.58–0.88 (strict vs. loose) (Baltra et al., 17 Jan 2026).
• Chiloe vs. Trinocular: High agreement on true positives; most false positives are persistent VP peninsulas, few false negatives due to short duration events.

Independence of VPs is quantitatively confirmed (pairwise similarity $< 0.14$). Event analyses show that three geographically diverse VPs suffice for stable detection rates (Baltra et al., 2024, Baltra et al., 17 Jan 2026).

5. Quantitative Findings on Routing Transients

Core metrics reveal that routing transients contribute significantly to observed Internet instability:

• Peninsula Frequency: Peninsulas are at least as prevalent as full outages, occupying a time fraction $\sim7.5 \times 10^{-4}$ of block-time in Trinocular (2017Q4). Outages are comparably rare.
• Duration Distributions: Approximately 65% of peninsulas last 20–60 min, characteristic of routing transients. Only 2% persist ≥24h, yet these outlier events account for 57% of total peninsula-time. The persistent fraction (≥5h) comprises 7% of all events but 90% of total peninsula-time (Baltra et al., 2024, Baltra et al., 17 Jan 2026).
• Spatial Scope: Around 10% of peninsulas affect an entire prefix, indicating inter-domain (policy) causes. Nearly 33% impact <1% of a prefix, suggestive of intra-AS or local disruptions.
• Island Rarity: Islands occur at a rate of ~7.7 events/year per Trinocular VP and ~1.75/year per Atlas VP, typically lasting <3,000 s and rarely involving more than one network hop.

6. Operational and Policy Implications

Recognizing routing transients is essential for both operational fidelity and governance:

• Measurement Noise: Islands and peninsulas contribute noise 5–9.7 times greater than actual operational events in systems like DNSmon. Filtering these phenomena clarifies true service failures (Baltra et al., 2024, Baltra et al., 2024).
• Misclassification Risks: Outage monitors that do not distinguish peninsulas risk false positives and obscured operational signals.
• Policy Impact: Quantitative detection frameworks inform debates on Internet fragmentation, "secession," sanctions, and peering disputes. No single country or RIR controls >50% of active addresses, precluding unilateral core reformation.
• Architectural Insights: Reachability-based definitions provide a robust test for transition planning (e.g., IPv6 must form its own >50% SCC to supplant IPv4).
• Mitigating Fragmentation: Persistent peninsulas reveal lasting policy-driven segmentation (e.g., Cogent vs. HE IPv6 dispute), while transient events highlight areas susceptible to operational remediation.

7. Broader Significance for Internet Health and Evolution

Routing transients, as measured by peninsulas, are not sporadic anomalies but an endemic feature of the post-global Internet (Baltra et al., 17 Jan 2026). Their explicit detection enables:

• Higher-resolution outage classification and reduced monitoring noise.
• Enhanced ability to trace the root causes of partial reachability—ranging from misconfiguration to macro-scale peering disputes.
• A principled connectivity-based framework for evaluating the Internet’s resilience, susceptibility to policy-induced partition, and prospects for seamless end-to-end innovation.

Making partial reachability as measurable and routine as full outages lays the groundwork for more accurate network health metrics and more informed policy regarding sovereignty, fragmentation, and future architectural shifts (Baltra et al., 2024, Baltra et al., 2024, Baltra et al., 17 Jan 2026).