39건의 기록된 인시던트

글로벌 인터넷 장애의
완전한 역사

1980년 ARPANET 붕괴부터 오늘날의 해저 케이블 절단까지 — 복원력 있는 네트워크 구축에 기여한 인시던트의 연대기.

BGP 인시던트 DNS 장애 케이블 절단 클라우드 장애 DDoS 공격 셧다운
스크롤하여 탐색
2020s
Dec 5, 2025

Cloudflare brief global disruption from configuration error (reported)

≈25 minutes (reported)
영향
Global
근본 원인
Human error, Misconfiguration
Bad config deployment affected request handling; rollback resolved quickly (reported).
"Short-lived outages/errors across many proxied sites."
CDN/Cloud services
Oct 19, 2025

AWS us-east-1 major cascade outage (DynamoDB/EC2) (reported)

≈15 hours (reported)
영향
Global (services depending on us-east-1)
근본 원인
Software bug, Systemic cascade
Prior response claimed DynamoDB DNS/discovery issue cascaded into EC2 control plane disruption; not verified here.
"Large downstream app outages and delayed autoscaling/instance launches (reported)."
Cloud provider
Sep 6, 2025

Red Sea cable cuts (SMW4 & IMEWE) (reported)

Weeks (repairs)
영향
Middle East, South Asia, East Africa, Europe–Asia transit
근본 원인
Unknown, Possible sabotage, Human error
Multiple cables cut; attribution unclear.
"Regional slowdowns and rerouting/latency increase."
Submarine cables
Jul 24, 2025

Starlink global outage (core software services failure) (reported)

≈2.5 hours (reported)
영향
Global Starlink footprint
근본 원인
Unknown, Software bug (suspected)
Reported failure of internal software services operating the core network (provider statement reported by analysts).
"Connectivity loss for satellite users."
Satellite ISP (LEO)
Forrester Kentik
Jun 12, 2025

Cloudflare outage attributed to Workers KV incident (reported)

≈2.5 hours (reported)
영향
Global (Cloudflare customers)
근본 원인
Software bug, Change error
Prior response claimed Workers KV cascade; not verified here.
"Customer API/control plane disruption; some site/app errors."
CDN/Cloud services Control plane / distributed storage
May 20, 2025

Global routing instability from malformed BGP attribute (Prefix-SID) (reported)

Minutes (reported)
영향
Global (multi-AS)
근본 원인
Software bug, Misconfiguration
Malformed BGP attribute propagated; certain vendors allegedly crashed/reset sessions.
"Brief reachability issues for some networks; high routing churn."
BGP routing Router software interoperability
Jan 15, 2025

Targeted cable incidents around Taiwan (reported)

Weeks for local restoration (reported)
영향
Taiwan and offshore islands (reported)
근본 원인
Possible sabotage, Hardware/physical failure
Incidents attributed to vessel activity; intent disputed.
"Local outages and increased geopolitical concern."
Submarine cables
Nov 17, 2024

Baltic Sea subsea cable disruptions (reported sabotage)

Repairs over weeks (reported); immediate impact rerouted
영향
Northern Europe (Finland, Sweden, Baltics, Germany links)
근본 원인
Unknown, Possible sabotage
Near-simultaneous cable cuts raised suspicion; attribution uncertain.
"Reduced capacity and higher latency; national security concern."
Submarine cables
The Guardian AP News
May 12, 2024

East Africa subsea cable damage (SEACOM & EASSy) (reported)

Weeks (repairs into early June)
영향
East and Southern Africa
근본 원인
Human error, Hardware/physical failure
Anchor drag / storm-related incident damaged multiple cables.
"Major slowdowns and partial outages; increased latency; business disruption."
Submarine cables
Internet Society Kentik
Feb 9, 2024

Red Sea subsea cable cuts linked to Rubymar (AAE-1 / EIG / SEACOM) (reported)

Weeks (repairs)
영향
Middle East, South Asia, East Africa, Europe–Asia transit
근본 원인
Human error, Conflict-related incident
Anchor drag from damaged/drifting vessel reportedly severed multiple cables.
"Regional slowdown/outages; rerouting and latency spikes."
Submarine cables
CSIS
Jan 3, 2024

Orange Spain routing registry compromise / BGP disruption

Hours (reported)
영향
Spain (primarily)
근본 원인
Cyberattack / credential compromise
RIPE registry account compromise enabled route origin record changes causing routing issues.
"Connectivity disruption for ISP customers and hosted services."
BGP routing IRR/RPKI management
APNIC Blog SecurityWeek
Nov 2, 2023

Portland (Flexential) data center power failure impacts Cloudflare & Workday

Hours to ~1 day+ for full restoration (varied by service)
영향
Global services with dependencies; physical location Portland, OR (USA)
근본 원인
Hardware failure
Power system failure and cascading generator/utility issues caused facility-wide outage.
"Loss of visibility/control for customers; SaaS downtime for enterprises."
Data center/power infrastructure Cloud/CDN control-plane
Cloudflare Cloudflare
Jun 13, 2023

AWS us-east-1 outage (Lambda-related cascading impacts)

≈2 hours (reported)
영향
Global (apps depending on us-east-1)
근본 원인
Software bug
Lambda capacity-management bug triggered at scale, affecting dependent AWS services.
"App outages and degraded performance across multiple brands."
Cloud provider Serverless platform
May 9, 2023

Pakistan mobile internet shutdown (political unrest)

≈3 days (reported)
영향
Pakistan (nationwide)
근본 원인
Political or regulatory action
Government ordered mobile networks offline; social platforms blocked during unrest.
"Major societal and economic disruption; suppressed information flow."
Government shutdown Mobile data
Internet Society Pulse Al Jazeera
Apr 4, 2023

Virgin Media UK ISP outages (BGP withdrawals/internal failure)

Two major outages (hours each) reported same day
영향
United Kingdom
근본 원인
Unknown, Routing/control-plane failure (suspected)
Routes withdrawn and traffic dropped at edge; internal failure led to customer disconnects.
"Nationwide consumer/business connectivity disruption."
ISP BGP routing
Feb 2, 2023

Matsu Islands (Taiwan) dual undersea cable cuts

Weeks (reported)
영향
Matsu Islands (Taiwan)
근본 원인
Hardware/physical failure, Possible sabotage
Two cables severed in early February; attributed to vessel activity (disputed intent).
"Severe local connectivity loss; reliance on microwave/satellite fallback."
Submarine cable
Jan 25, 2023

Microsoft Azure & Microsoft 365 global outage (WAN routing incident)

Several hours (reported)
영향
Global
근본 원인
Misconfiguration, Change error
Microsoft WAN routing change caused instability/route flaps impacting Azure and M365 services.
"Cloud and productivity suite disruptions affecting organizations worldwide."
Cloud provider WAN routing/BGP
ThousandEyes Practical365
Oct 25, 2022

Sudan protest-related national internet shutdown

≈8 hours (reported)
영향
Sudan (nationwide)
근본 원인
Political or regulatory action
Authorities reduced/disabled connectivity during protests.
"Information suppression and disruption to services."
Government shutdown
Oct 19, 2022

Shetland Islands submarine cable cut (SHEFA-2)

≈14.5 hours (reported)
영향
Shetland Islands, Scotland (UK)
근본 원인
Human error, Hardware/physical failure
Cable severed (often attributed to anchor/fishing activity).
"Local isolation; emergency services needed backups."
Submarine cable
The Guardian Shetland News
Sep 1, 2022

Iran protest-related internet shutdowns (curfews and regional blackouts)

Recurring daily hours + multi-day regional events
영향
Iran (nationwide and regional)
근본 원인
Political or regulatory action
Authorities restricted/disabled mobile internet and platforms during protests.
"Communication suppression and economic disruption."
Government shutdown Mobile networks
Jul 8, 2022

Rogers Communications nationwide outage (Canada)

≈19–24 hours (major disruption)
영향
Canada (nationwide)
근본 원인
Human error, Misconfiguration
Maintenance change removed a routing filter and triggered cascading routing/control-plane failures.
"Payment disruptions, emergency-call impacts, widespread societal disruption."
ISP/telecom core
ThousandEyes Government of Canada / CRTC
Oct 4, 2021

Facebook/Meta global outage (BGP + DNS reachability)

≈6 hours
영향
Global
근본 원인
Human error, Misconfiguration
Backbone maintenance change withdrew BGP routes; DNS and internal tooling became unreachable, complicating recovery.
"Global disruption of messaging/social communications; business disruptions."
BGP routing DNS Private backbone
Engineering at Meta Cloudflare
2010s
Jun 24, 2019

Verizon BGP route leak impacting Cloudflare (and others)

≈1–2 hours
영향
Global
근본 원인
Misconfiguration, Filtering failure
Customer route leak propagated through Verizon due to insufficient filtering; misrouting/blackholing ensued.
"Partial unreachability/slowness for many services; routing community alarm."
BGP routing
Cloudflare Cloudflare
Feb 28, 2017

AWS S3 outage (us-east-1) from operator error

≈4 hours
영향
Global (customers relying on us-east-1)
근본 원인
Human error, Operational tooling/controls gap
Incorrect parameter in internal tool took down more S3 subsystems than intended; slow index recovery.
"Major web/app disruptions due to reliance on S3 in a single region."
Cloud provider Storage service
AWS The Guardian
Oct 21, 2016

Dyn DNS DDoS (Mirai botnet)

Intermittent over the day (multiple waves)
영향
North America, Europe, Global users of affected sites
근본 원인
Cyberattack / DDoS
IoT botnet traffic overwhelmed Dyn’s DNS infrastructure.
"Widespread inability to resolve domains for major services."
DNS DDoS
ThousandEyes KrebsOnSecurity
Jan 27, 2011

Egypt nationwide internet shutdown (Arab Spring)

≈5 days (major blackout)
영향
Egypt (nationwide)
근본 원인
Political or regulatory action
Government ordered ISPs to disconnect and withdraw routes to suppress protests.
"Severe societal and economic disruption; communications blackout."
Government shutdown BGP withdrawals
Wikipedia (overview) RIPE NCC (related analyses via RIPE Labs/archives)
Apr 8, 2010

China Telecom large-scale route leak/hijack (reported)

≈18 minutes (often reported)
영향
Global
근본 원인
Misconfiguration, Disputed
Large number of prefixes announced via China Telecom; traffic detoured. Intent disputed.
"Potential traffic detours and risk of interception; user-visible impact unclear."
BGP routing
2000s
Feb 24, 2008

YouTube global outage via Pakistan Telecom BGP hijack

≈2 hours
영향
Global
근본 원인
Political or regulatory action, Misconfiguration, Upstream filtering failure
Domestic blocking route advertised globally due to upstream propagation and insufficient filtering.
"Global unreachability of YouTube during incident window."
BGP routing Policy-driven filtering leak
Google Research CircleID
Jan 30, 2008

Mediterranean & Gulf submarine cable cuts (Middle East/South Asia disruption)

Days to weeks (repair dependent)
영향
Middle East, North Africa, South Asia (multi-country)
근본 원인
Hardware/physical failure, Human error, Unknown
Multiple cable faults/cuts; some attributed to anchor drag, some unclear.
"Severe bandwidth reduction and international connectivity degradation."
Submarine cables
RIPE NCC Wikipedia
Dec 26, 2006

Taiwan (Hengchun) earthquake submarine cable breaks

Weeks (full repair reportedly ~49 days)
영향
East Asia / Asia-Pacific (Taiwan, China, Hong Kong, Japan, Singapore, etc.)
근본 원인
Natural disaster
Undersea earthquake and landslides severed multiple major cables near Taiwan/Luzon Strait.
"Severe international connectivity degradation; high latency due to reroutes."
Submarine cables International backbone
Submarine Networks International Cable Protection Committee (PDF)
Jan 25, 2003

SQL Slammer worm

Hours (peak), with lingering effects over 1–2 days
영향
Global (notably Asia, North America, Europe)
근본 원인
Cyberattack / malicious code, Software vulnerability
Worm exploited MS SQL Server vulnerability; explosive scanning caused congestion and outages.
"Severe congestion and widespread service disruptions."
Internet Cyber incident / traffic flood
NETSCOUT ASERT WeLiveSecurity (ESET)
1990s
May 19, 1998

Galaxy IV satellite failure

Days to ~1 week (varying by service)
영향
United States (nationwide), some Canada
근본 원인
Hardware failure
Satellite control failure caused loss of service for pager/broadcast transponders.
"Mass pager outage affecting businesses and emergency communications."
Satellite communications
The Atlantic National Academies
Apr 13, 1998

AT&T frame relay cloud outage

Up to ≈26 hours for some customers
영향
United States (nationwide corporate WAN customers)
근본 원인
Software bug, Change error
Software upgrade failure and error flooding caused network-wide instability.
"Corporate WAN disconnects; disrupted financial and airline operations (reported)."
Enterprise data network (Frame Relay)
Wired Los Angeles Times
Jul 17, 1997

“.com/.net DNS” outage (Network Solutions zone file error)

≈4 hours
영향
Global
근본 원인
Human error
Corrupted/incorrect zone file update disrupted resolution for large numbers of domains.
"Large-scale name resolution failures; widespread site/email disruption."
DNS (TLD name servers)
BYTE (PDF) squid-cache.org
Apr 25, 1997

AS7007 BGP route leak

Hours (variously reported; not precisely captured here)
영향
Global
근본 원인
Human error, Software bug, Misconfiguration
Misconfigured router and bug leaked more-specific routes and hijacked large portions of routing table.
"Widespread reachability problems due to misrouting/blackholing."
Internet backbone BGP routing
Wikipedia Cloudflare (PDF)
Jan 4, 1991

New York fiber cut disrupts multi-state communications

Hours (same-day restoration reported)
영향
US Northeast (New York, DC, Boston cited)
근본 원인
Human error, Hardware/physical failure
Accidental severing of a major fiber trunk (construction/backhoe-style incident).
"Call blocking and disruptions to trading/aviation communications (reported)."
Fiber backbone (telecom/data)
Jan 15, 1990

AT&T long-distance network crash

≈9 hours
영향
United States (nationwide long-distance)
근본 원인
Software bug
Faulty switch software update triggered cascading switch resets and signaling storms.
"Large-scale call failures; business disruption; delayed flights reported."
Telephony (PSTN long-distance) Switching network
Cal Poly (jdalbey) case study The Software Horror Picture Show (Uni Saarland)
1980s
Nov 2, 1988

Morris Worm outbreak

Days (major disruption over multiple days)
영향
Primarily US academic/military networks, Spillover impacts globally
근본 원인
Cyberattack / malicious code, Software vulnerability, Human error
Self-replicating worm exploited weak passwords and vulnerabilities, saturating systems and networks.
"Severe network congestion and system outages; significant cleanup costs."
Internet (early TCP/IP) Cyber incident
Wikipedia CERT/CC (SEI)
Oct 27, 1980

ARPANET collapse (routing/IMP bug cascade)

≈4 hours
영향
United States (ARPANET nodes)
근본 원인
Software bug
Status-message propagation and IMP error-handling/garbage-collection bug caused network-wide disruption.
"ARPANET connectivity loss across research sites."
Early Internet (ARPANET) Routing infrastructure
Computer History Museum VICE
2025 Cloudflare brief global disruption from configuration error (reported)

데이터에서 보이는 패턴

수십 년간의 네트워크 장애는 인터넷이 어떻게 고장나는지에 대한 반복되는 주제를 보여줍니다.

가장 흔한 원인
Human error
13건의 인시던트
최다 발생 10년
2020s
22건의 기록된 인시던트
총 기록 건수
39
주요 인시던트
기간
1980 – 2025
45년의 역사

역사가 가르치는 것

모든 대규모 장애는 더 복원력 있는 시스템 구축을 위한 교훈을 남깁니다.

인적 오류가 공통 원인

1990년 AT&T 장애부터 2021년 Facebook 장애까지, 잘못된 설정과 운영 오류는 다른 어떤 요인보다 더 많은 장애를 일으킵니다. 현대 네트워크의 복잡성은 단 하나의 오타가 전 세계로 확산될 수 있음을 의미합니다.

BGP는 여전히 인터넷의 아킬레스건

네트워크 간 트래픽을 라우팅하는 Border Gateway Protocol은 암묵적 신뢰 시대에 설계되었습니다. 수십 년이 지난 지금도 RPKI와 같은 보안 조치의 도입이 불완전하여 경로 유출과 하이재킹이 계속되고 있습니다.

물리적 인프라는 놀라울 정도로 취약

해저 케이블은 대륙 간 데이터의 95%를 전달하지만, 닻, 지진, 그리고 점점 증가하는 의도적 방해에 취약합니다. 홍해와 같은 지리적 병목 지점은 여전히 중대한 장애점입니다.

집중화가 시스템적 위험을 만든다

AWS us-east-1이 다운되면 인터넷의 절반이 영향을 받습니다. Cloudflare에 문제가 발생하면 수백만 개의 웹사이트에 접속할 수 없게 됩니다. 중앙 집중화로 얻는 효율성에는 과도한 장애 모드가 따릅니다.

다음 장애가 영향을 미치기 전에 대비하세요

Latency Global은 70개 이상의 글로벌 위치에서 서비스를 모니터링합니다. 고객 불만이 아닌, 문제 발생 시 알림을 받으세요.

무료 체험 시작

$5/월 • 약정 없음 • 언제든 취소 가능