The aws deployment system notices, when resources become too limited and expands the network with the necessary additional machines that fit the demand (GPU, RAM...)
It also gives back resources that are no longer needed
When a user or a service (e.g., an iterator service) knows in advance that he will be requiring a certain amount of dedicated resources, those resources can be reserved. (For now, this functionality does not need to be interfaced to normal platform users.)
User story:
A parallelized series of EM simulation instances using GPU-accelerated FDTD is set up using the sweeper functionality.
The cluster notices that there are not enough gpu machines available to handle the parallel load and adds more machines.
Those machines are released again, when the system notices that they are no longer needed
Dynamic computational resource allocation, scalability
Definition of Done:
User story: