Ask HN: Using AWS to Approximate an Academic HPCC

In a typical academic/super computer set up, multiple useres are allowed to ssh into some head node, and there is a job scheduler installed that allows you to submit jobs requiring some resources. It then farms the jobs out to other compute nodes.

Is it possible to approximate this workflow using AWS, where you ssh into a very cheap always on node like a t1.micro, but then from the command line can spin up another ec2 instance that runs a script or pipeline, saves the output to S3, and terminates itself?