Since the bebopd and PCT are portals processes, they must not be started on a machine until the system is ready to register portals processes and handle incoming and outgoing portals messages. Currently, since Cplant runs Linux and sends portals traffic over Myrinet, this means that the MCP (Myrinet Control Program) is loaded on the LANai network card and the portals module (p3mod), the Cplant module (cTask), and Myrinet driver module (rtscts) are installed.
The bebopd is a single process that runs on a service node and maintains the state of the compute partition. As such, the bebopd is synonymous with a virtual machine. If two bebopd daemons are running on two different service nodes, managing two disjoint sets of compute nodes, then logically the Cplant appears to be two separate clusters. There is nothing physically preventing communication between service and compute nodes of the two machines, but due to the way bebopd administers it's partition, it won't happen.
In fact, Cplant can be partitioned into virtual machines, each consisting of a collection of compute nodes and service nodes, each with it's own version of Cplant software and configuration files. The directories and scripts on the administrative network that make this happen are described elsewhere. The point here is that one bebopd will be running on each virtual machine. Users logging in to a service node of a given virtual machine will receive compute resources allocated by the bebopd running on that virtual machine.