PCCM2 supports restart and branch runs. History regeneration runs are not supported. The difference between a restart and a branch is that a restart will number subsequent history and restart files as if the run had simply continued from the point of the restart. A branch, on the other hand, starts numbering of the history and restart files at the beginning. Because all model settings come from the restart files and from the namelist files (not from previously written history files or regeneration files), there is no other substantive difference between a restart run and a branch run with the parallel model.
A restart run is specified by setting the namelist variable NSREST to 1 (one). A branch run specified with a value of 3 (three). (For an initial run, the setting of NSREST is 0 (zero)). If a restart or branch is specified, the model opens the Master and Primary restart data sets whose base name is specified by the string-valued namelist variable NREVSN. For example, to restart from the files r0009 and r0009.A, specify
NSREST = 1,
NREVSN = 'r0009',
The model can be restarted directly from the per-processor restart files (r0009.A.0000, r0009.A.0001, and so on) provided the restart run is on the same number of processors as the original run. Adding `*' to the value of NREVSN
NREVSN = 'r0009*',
indicates that you are restarting from the file r0009. The Intel PFS restart file is also specified with a * even though it is a single file. This is because it is written from all nodes simultaneously. When restarting on a number of processors different from the original run, you must combine the Primary Restart Data in a single file (see above) and leave the `*' off the NREVSN string. If NSREST is 0 (zero) for an initial run, the setting of NREVSN is ignored.
Restart files generated by CCM2 cannot be used to restart PCCM2 or vice versa. The structure of the records in PCCM2 is vertical column oriented, with all fields associated with a column of grid points grouped together on output. Since restarting on a different number of processors will involve a different allocation of columns to processors this allows efficient, parallel input of the checkpoint/restart data.