Support scheduling systems

[inJeans]

Is it possible to integrate DIGITS with a cluster schedueler like slurm?

For example, if you could have a local instance of the digits server running and then submit a job to the queue for training.

Similar to the way nvvp handles remote nodes.

[lukeyeager]

We briefly considered using a tool built on top of nvidia-docker but had to abandon it for other priorities.

[lukeyeager]

See related discussion at #108.

[TimZaman]

A message queueing system might be nice indeed (rabbitmq/zmq), where workers can register themselves be assigned jobs. But that could turn into a docker bonanza and setting it up would not be as turn-key as it is now.