Next: Sub-clustering
Up: Designing the system
Previous: The user's view
  Contents
The scheduler
``Scheduler'' is not an entirely covering name for this central part
of the system. In fact, the scheduler is supposed to take care of
several tasks, that make this system look as one large machine:
- Receive jobs (pieces of serial code) from clients
- Parallelize code wrt. the current state of the cluster
- Schedule and distribute code and data to selected nodes
Right now, the scheduler does not take requests from clients (since
there are no clients). The scheduler is run with a program name as a
command-line argument, and will then load that program into the TONS
system, and perform parallelization and scheduling on that
program. When the nodes return the result, the scheduler
terminates. This is something that will change, but the current
behavior is sufficient for development.
1999-08-09