We just launched AWS Global Accelerator! aws.amazon.com/blogs/aws/new-… With it, you can use our global backbone to accelerate your applications. Now, allow me to nerd out and go deep on how some of the fault tolerance actually works ...
Global Accelerator uses BGP anycast. We give you an IP address that we announce from lots of locations on our backbone and "The Internet" figures out getting your users to the closest site.
Until now we've haven't offered anycast services, because fault tolerance is very very hard. We actually tried it for CloudFront years ago, but found it wasn't reliable or precise enough. So what's new?
The core challenge with anycast is that when a link dies, the networks connected to that link think that it's still where they should be sending packets. For a while at least ... up to minutes. This can cause congestion and black-holing ... which isn't, you know, good.
With Global Accelerator we give you what we've been doing for Amazon Route 53 for 8 years now ... striped, resilient, IPs. Here's what that means: you can have multiple static IPs .. and they are meaningfully redundant. We don't advertise these IPs on the same links!
So if there's a problem with a link, and downstream networks aren't caught up yet, only one IP should be affected. The other will be fine. You can still hardcode both IPs, they're both static and just for you, and ordinary client-level retries and resilience will take care of it.
This kind of "striping" for an anycast service is very very reliable, and much better plain anycast. I love seeing how our customer's priorities (availability in this case) end up being reflected in our products. It'll be awesome to see what applications customers build! FIN.
