Say I have a task which can be ideally parallelized and requires 600 CPU-minutes to finish.
- I can complete it on my workstation in 10 hours.
- I can launch 10 EC2 instances and complete it in 1 hour.
- Is there a way I can run it on 100 machines and have it finish in 6 minutes? (and not pay for more)
- 1000 machines and have it finish in 36 seconds?
Perhaps AppEngine-MapReduce? Is there anyone else selling CPU time in units smaller than 1 hour?