What's the best way to provide continuous access to AWS-based apps in the event of Amazon downtime?

While not direct answers, the blog posts on SLAs and availability from earlier this year might be useful:
Service Level Agreements, pt. 1
Service Level Agreements, pt. 2


b

The techniques for failing over are fundamentally unchanged, clouds or no clouds.

I don't know of any services that provide AWS compatibility, as vendors are hesitant to implement the API. (There are API compatible implementations with things like Eucalyptus/Ubuntu, just bring your own webops.)

Five nines is 5.26 minutes of downtime per year. To guarantee that level of service the fail over strategy better not involve bringing up new virtual machines. If you don't have redundant systems that are always up, five nine is not happening. That's before even thinking about the implications of what it means for the service to be 'up' or what might be the harder problem, dealing with the data.

I think it is worth noting that every 9 is a factor of 10 better service level, and may cost a factor of 10 or more to guarantee in some cases. Few organizations have five nines as a business 'requirement'. More nines are nice to have but pursuing them may provide diminished returns from an ROI perspective.

The cost of a fully replicated application on standby in an alternate cloud is probably hard to justify. Being able to bring the application up ASAP (on the order of minutes) on a new provider is probably sufficient for most organizations and is largely the same work that needs to be done to provide predictability and consistency. (automated provisioning, configuration management and orchestration)

In my opinion cloud computing doesn't change the fundamentals so much as collapse the time scale. What used to require people doing and maybe hours of work can be done in minutes with APIs.

I'd also love to hear what people have done.

The space will be interesting to watch as vendors emerge and consolidate, but off the top of my head the usual suspects after AWS are GoGrid, Flexiscale and Rackspace. What else are people using? I'd love to hear that too.

Tomorrow's datacenter is clearly API driven compute, storage and network resource abstractions.