Cplant runtime components

The bebopd node allocator has been enhanced to support PBS operation on Cplant. In particular:

• The bebopd node allocator daemon (when run in PBSsupport mode) ensures that yod requests from a PBS job do not ask for too many nodes. It determines how many nodes a PBS job is already using before granting more nodes to the yod request.

• The bebopd knows how many live, working compute nodes there are in the virtual machine. The PBS server has an attribute called resources_available.size which it hopes is this value. The bebopd (when run in PBSupdate mode) sets this value in the server and keeps it up to date when it changes.

• The bebopd can be configured to reserve some number of nodes for non-PBS jobs. See the PBSinteractive options of bebopd and pingd. Users can acquire these nodes for their application simply by running yod on a service node.

Note that the bebopd does not reserve specific nodes for non-PBS jobs. It just ensures that the non-PBS jobs are not using more than the specified number of nodes. Also, PBS does not allocate specific nodes to PBS jobs, it allocates them some number of nodes. The bebopd selects the particular nodes that will be allocated to yod jobs started by a PBS job.

pingd has been enhanced as well:

• The pingd utility has been enhanced to list the PBS job ID (if any) of the Cplant jobs it displays. Each PBS job may start more than one Cplant job.

• pingd may also be used to alter the bebopd's support of PBS while the bebopd is running. See the -PBSsupport, -PBSupdate, and -PBSinteractive options of pingd.

• pingd also allows you to specify a PBS job ID on it's command line. It will list every Cplant job that is part of that PBS job.

• The summary information provided by pingd mentions how many nodes are running PBS jobs, how many are running interactive jobs, and how many are reserved for interactive jobs.