Bangladesh internet shutdown revealed in BGP data

In July 2024, Bangladesh’s internet connectivity collapsed after operators withdrew BGP prefixes in a coordinated sequence. Public routing data and an operator’s first-hand account reveal how the shutdown unfolded, offering lessons for network resilience and monitoring worldwide.

The outage began on 17 July 2024, when Grameenphone (AS24389) withdrew its 138 prefixes, followed by Banglalink (AS25245), Robi Axiata (AS24432), Earth Telecom (AS58715), and Carnival Internet (AS63526). By 18 July, 170 million people had lost access, and the country disappeared from the global routing table. The withdrawals were manual, session-by-session, with no automation distinguishing the shutdown from routine maintenance.

How the shutdown unfolded

The withdrawals occurred in a staggered sequence over roughly 20 hours, with no shared upstream failure visible in the routing data. Grameenphone’s prefixes vanished first, followed by Banglalink, Robi Axiata, Earth Telecom, and finally Carnival Internet. The operator of AS63526, who documented the process, noted that the commands used were identical to those for routine maintenance—no protocol or tooling flagged the event as extraordinary.

A notable anomaly emerged in the data: eight of Grameenphone’s IPv6 prefixes remained in the global routing table for 16 hours after the main withdrawal. This suggests either a missed manual checklist or a partially automated process requiring follow-up. The restoration process was equally uneven, with broadband services returning before mobile networks, and route flapping visible between waves of re-announcements.

What the data reveals—and what it doesn’t

Public routing data from RIPE RIS, Cloudflare Radar, and IODA confirmed the outage’s scale and timing, but it cannot explain the motivations behind the withdrawals. The operator’s account fills some gaps, showing how manual processes left traces in the routing table. However, the data does not clarify whether the shutdown was centrally coordinated or whether operators acted independently under government orders.

IODA’s three independent signals—BGP routing, active probing, and darknet traffic—all dropped to near-zero simultaneously, ruling out measurement artifacts. The restoration’s asymmetry, with broadband returning before mobile networks, further suggests deliberate prioritization rather than a technical recovery sequence.

Lessons for network operators

The Bangladesh shutdown highlights vulnerabilities in global routing infrastructure. Despite high RPKI ROA coverage (98% in Bangladesh), ROV enforcement at internet exchange points (IXPs) cannot prevent operators from withdrawing their own prefixes. The event also underscored the limitations of RIPEstat’s 8-hour observation windows, which are too coarse for real-time monitoring. RIPE RIS Live, which streams BGP updates in real time, would have provided finer-grained visibility.

Local peering also proved critical. Networks that peered through BDIX, Bangladesh’s domestic IXP, maintained internal connectivity even as international transit links went dark. This resilience layer, however, was limited—only 7.6% of Bangladesh’s networks peer locally, leaving most traffic dependent on international transit.

Sources

Inside a Shutdown: BGP Evidence from an Operator Who Was There – RIPE Labs