Life of a typical computing job (2)

• The script is run remotely on one of the reserved compute nodes, a program 'mpirun' starting the program as a process on each of them. Each process gets a unique identity, when running the MPI initializing subroutine/function call

• Each process communicates with the other processes, reading and writing in local and global files

• The parallel program exits normally, exits because of a serious error condition, or is aborted by the batch system because it has reached the maxjoblength limit

• All the reserved compute nodes are cleaned out automatically, removing still existing processes and local files. A log file, the global files and the home directory files are kept. An e-mail is sent to the user.