There is only one master task in the virtual machine. It is responsible primarily for task scheduling and result collecting. The master task is run directly from the GUI, but all other PVM tasks are spawned by the master task itself.
Once the number of hosts running in the virtual machine has been determined, the master task spawns servant tasks to perform the jobs defined by the user in the GUI. Because the master task spends most of its time in a blocking wait state for the servant tasks to complete their jobs, a servant task also runs on the master node to increase the overall throughput.
Once the servant tasks have performed all of the desired simulations, a send data task is spawned by the master on each host. After the results are all back on the host node, a single SAS and graphics output file is generated for each scenario from the many independently generated statistics files. These files along with any saved land cover maps are then installed in the permanent GIS. This final step is cumbersome and costly in terms of communication bandwidth, but the results need to be centrally stored in order for them to be of any use.