It would be neat to enable the system to be run as a cluster. Clients could then submit jobs to the cluster without needing to know which machine was actually running their job.
There are obviously some complex issues here:
- how do requests get routed to worker nodes? If a job is running on one host a status request for that job also needs to go to that host.
- do we need load-balancing?
It would be neat to enable the system to be run as a cluster. Clients could then submit jobs to the cluster without needing to know which machine was actually running their job.
There are obviously some complex issues here: