It has happened again, one (albeit large) provider has a problem, and it takes down vast quantities of the internet, and takes hours, if not longer to get it all back to stability. Does this show that the original plans and early design ideas of the internet being a decentalised system are now a thing of the past?
Quite possibly, with the services provided by hyper-scalers – AWS, Azure, GCP, Cloudflare and the like, more and more services are being brought together rather than separated out.
What Happened?
Monday morning started with problems localised to the US-East-1 Availaiblity Zone within AWS, this is the oldest and default/primary zone for a lot of services. This includes Amazon services themselves, so when there was a problem it had a knock on effect taking more out and causing further problems. The actual initial problem was resolved within a few hours but it took at least 15 hours for most systems to come back to normal stability.
Decentralisation
When the internet was initially designed, it was set up to be decentralised, with everyone able (and encouraged) to use multiple providers, to spread out the resources, to use separate systems. While this has for a long time been the case, with the corporate nature of changes and the interest in centralisation and profits, providers became more keen to cover more and more solutions. From there hyperscalers were born. There’s nothing you could want to do that can’t be run on one of these solutions, but is it a good plan? Maybe not, as shown over the past few days how many systems were broken or outright down for long periods of time.