Maybe you could even timeshare your machines out for some other instances for a discount on your service, similar to how you can send solar power back up the power grid for an electricity credit. Then ec2 really WOULD be a cloud!
Amazon could even simply benchmark your server and network connection, and compensate you as a function of that, which would drastically reduce the incentive to hook up old crappy hardware to the cloud.
It's really the idea of being able to run copies of your OWN instances locally that intrigues me though.
No reason you can't already do this.
Just rather than supplying the hardware you have to rent it from Amazon.
Would be interesting to know how these instances were affected by the outage. Since the problem was with the EBS infrastructure not the actual compute hosts I'm guessing this wouldn't have helped any
"Eucalyptus is a software platform for the implementation of private cloud computing on computer clusters. There is an enterprise edition and an open-source edition. Currently, it exports a user-facing interface that is compatible with the Amazon EC2 and S3 services but the platform is modularized so that it can support a set of different interfaces simultaneously. "
One of the features that allows AWS to scale is that it runs on a unified and consistant hardware platform. Take a look at the recent Google Data Center video or the Facebook Open Server initiative - companies try where possible to run identical hardware for all of their needs as it removes inconsistances in behavior and performance between arbitrarily different pieces of equipment.
So the idea of giving Amazon your own hardware to host is terrible.
The other part of his idea sounds like mirroring servers so that if an EC2 instance goes down, some bare-metal hardware running elsewhere could kick in. That's totally feasible, but no need to do it in the same DC under Amazon's control. Just set up your own machines (bare-metal or virtual) and keep them in sync with your EC2 instance. In the event of EC2 going down you can just change your DNS over to point your fall-back IP or do some clever off-site IP load-balancing tricks (assuming the load balancer hasn't gone down too)
In fact this would have been a good strategy for the ECG monitoring folks if they needed to keep service up even during a prolonged EC2 outage.
I wonder if this will get more companies to maintain a non-cloud, vanilla setup on some dedicated or collocated boxes.
May be we should have an annual day to bring down our primary servers and see how the backups do. The idea shouldn't be to confirm if everything works or not. It should be to determine what did work and did not work.