The Sky Above the Clouds: A Berkeley View on the Future of Cloud Computing

The tl;dr I got out of the paper is they are calling for the creation of an intercloud broker layer that mediates between cloud vendors, and call it Sky Computing. I didn't see any extended discussion of how to work around egress charges designed to lock in customers, other than just implementing sharding with more steps: chunk of workload Foo is brokered to AWS, chunk of same type of workload but different data is brokered to GCP, and so on.

Another alternative is to deliberately design asymmetric data movement wherever possible: inbound into cloud vendors is nearly as-needed, outbound is miniscule and to something like Cloudflare R2. Or even minimize data movement at all. Those are unnatural contortions that are easily countered by different pricing policies by the cloud vendors though, which can lead to painful re-architecting and re-implementation.