In early 2024, African internet broke three times in three months. A thread on undersea cables, for people who think the cloud is in the sky (1/18)

1/ If you work remotely from Africa, your job is attached to a handful of cables lying on the sea floor. In 2024 we all learned exactly how few.

2/ February 2024: several cables in the Red Sea were damaged, including Seacom, AAE-1 and EIG. The damage was widely reported to have come from the dragging anchor of a cargo ship that had been attacked and abandoned. East African traffic to Europe and Asia had to be rerouted.

3/ 14 March 2024: four cables off the coast of Côte d'Ivoire went down on the same day: WACS, MainOne, SAT-3 and ACE. The cause was an underwater landslide in a deep canyon off Abidjan.

4/ Ghana, Côte d'Ivoire, Nigeria, Liberia, Benin and others saw major outages or heavy slowdowns. Even South African networks felt it, because some of those cables land there too.

5/ May 2024: EASSy and Seacom were both damaged off the coast of southern Africa. Kenya, Tanzania, Uganda, Rwanda, Mozambique, Malawi and Madagascar all felt it. Here in Nairobi, my company VPN took four minutes to connect for the better part of a week.

6/ So: three separate incidents in about three months, on both coasts. None of them was a hack. Anchors, landslides, the sea being the sea.

7/ Why does it take weeks to fix? There are very few cable repair ships, and the one that serves most of this coastline is based in Cape Town. It has to sail there, find the break, lift the cable from the bottom, splice it, and put it back. Weather matters. Permits matter.

8/ Why does one cut hurt so much? Because many "redundant" routes are not. Two ISPs that both buy capacity on the same cable are one ISP with two logos.

9/ The good news: new capacity has been landing. Google's Equiano and the Meta-led 2Africa cable both add routes on the west and east coasts. More paths means a single landslide is less likely to cut a country off.

10/ The bad news: landing a cable is not the same as reaching the town where you live. Terrestrial fibre, power, and the last mile are still the weak points.

11/ For remote workers: ask your ISPs which cables and which landing stations they use. Many will not know. The ones that do are the ones you want.

12/ Your mobile data is often NOT a real backup. Mobile operators ride the same international cables. Mobile is a backup for your local fibre being dug up, not for the sea.

13/ Agree with your employer, in writing and in advance, what happens on a cable day: async work, a shifted schedule, no penalty. The day you need that agreement is the worst day to negotiate it.

14/ For developers serving African users: if everything you have lives in one European region, a cable cut makes your app look broken even when it is fine. Cache aggressively, put static assets on a CDN with local presence, and make sure your error states say "slow connection" instead of "something went wrong".

15/ Watch your timeouts. A route that suddenly goes via Asia adds hundreds of milliseconds. Clients with 5 second timeouts start failing in loops, and retries make it worse. Exponential backoff, please.

16/ Status pages: if your status page is hosted in the same place as your app, it is decoration. And send status by SMS for the critical stuff. SMS kept working when the data did not.

17/ For policy people: cable repair capacity is infrastructure, like roads. A continent of 1.4 billion people should not depend on so few ships.

18/ Sources: Cloudflare Radar's write-ups on the March and May 2024 outages, which show traffic by country hour by hour; the operators' own statements at the time; TeleGeography's submarine cable map for routes and landing points. Pole to everyone who lost a client call that week. I lost two.