Are there any computation limits to creating a new test - i,e. flops, memory, etc.?
Yes. At the moment the OpenKIM system is limited to serial computations on a single processor. Although there is no formal limit on the amount of time calculations can take, they should be “reasonable”. However, we are actively working and collaborating with others to make parallel computations possible on OpenKIM.
What machine(s) are running these tests for all models? Is access to these machines provided by KIM or would the people uploading the test responsible for make arrangements to run the test for all models?
The computational resources are handled internally by OpenKIM and are opaque to the user. A user simply uploads a new Model or Test. The Processing Pipeline identifies the computational tasks to be performed and places then on a queue that is watched by a cloud-based network of “worker drones” that can then grab tasks and perform the calculations.
The worker drones are virtual machines that run on different computers on the web. Anyone interested in contributing idle compute cycles to the KIM project, can install a drone on their computer and become part of the KIM computational system – just let us know! (However this is not required to use KIM resources.)