Changes between Version 281 and Version 282 of doc/app/runtime_parameters


Ignore:
Timestamp:
Oct 3, 2017 11:36:45 AM (8 years ago)
Author:
raasch
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • doc/app/runtime_parameters

    v281 v282  
    175175Besides the standard quantities from the above list, the user can output any other quantities. These have to be defined and calculated within the user-defined code (see [[../userint/output|User-defined output quantities]]). They can be selected for output with the user-parameter [[../userpar#data_output_user|data_output_user]] for which the same rules apply as for '''data_output'''. Output of the user-defined quantities (time interval, averaging, selection of cross sections, etc.) is controlled by the parameters listed above and data are written to the same file(s) as the standard quantities.\\\\
    176176'''Output on parallel machines:'''\\\\
    177 By default, with parallel runs, processors output only data of their respective subdomains into separate local files (file names are constructed by appending the four digit processor ID, e.g. <filename>_0000, <filename>_0001, etc.). After PALM has finished, the contents of these individual files are sampled into one final file using the program {{{combine_plot_fields.x}}} (automatically activated by '''mrun''').\\\\
     177By default, with parallel runs, processors output only data of their respective subdomains into separate local files (file names are constructed by appending the four digit processor ID, e.g. <filename>_0000, <filename>_0001, etc.). After PALM has finished, the contents of these individual files are sampled into one final file using the program {{{combine_plot_fields.x}}} (automatically called by '''palmrun''').\\\\
    178178Alternatively, PALM is able to collect all grid points of a cross section on PE0 before an output is done. In this case, only one  output file ([[../iofiles#DATA_2D_XY_NETCDF|DATA_2D_XY_NETCDF]], etc.) is created and {{{combine_plot_fields.x}}} does not have to be called. In case of very large numbers of horizontal grid points, sufficient memory is required on PE0.  This method can be used by assigning [#data_output_2d_on_each_pe data_output_2d_on_each_pe] = ''.F.''.\\\\
    1791793d volume data output is always handled separately by each processor so that {{{combine_plot_fields.x}}} has to be called anyway after PALM has been finished.
     
    875875||''6'' ||same as ''4'', but with parallel I/O support ||
    876876'''Important:'''\\
    877 Setting netcdf_data_format > 2 requires a netCDF4 library (set -I, -L, and -l options for compiling and linking appropriately in configuration file). Also, preprocessor switch {{{-D__netcdf4}}} has to be set (see line starting with {{{%cpp_opts}}} in the '''mrun''' configuration file).\\\\
     877Setting netcdf_data_format > 2 requires a netCDF4 library (set -I, -L, and -l options for compiling and linking appropriately in configuration file). Also, preprocessor switch {{{-D__netcdf4}}} has to be set (see line starting with {{{%cpp_options}}} in the '''palmrun''' configuration file).\\\\
    878878Files with netCDF4 format cannot be read with netCDF3 libraries.
    879879
     
    11501150CPU time needed for terminal actions at the end of a run in batch mode (in s).\\\\
    11511151If the environment variable '''write_binary''' is set ''.T.'' (see [../runs Initialization and restart runs]), PALM checks the remaining CPU time of the job after each time step. Time integration must not consume the CPU time completely since several actions still have to be carried out after time integration has finished (e.g. writing of binary data for the restart run, carrying out output commands, copying of local files to their permanent destinations, etc.) which also takes some time. Furthermore, the parameter has also to account for the CPU time consumed by the job '''before''' the PALM executable has started (e.g. required for providing input data, or for compiling the user-interface code, etc.). The maximum possible time needed for these activities plus a reserve is to be given with the parameter '''termination_time_needed'''. Among other things, it depends on the number of grid points used. If its value is selected too small, then the respective job will be prematurely aborted by the queuing system, which may result in a data loss and will possibly interrupt the job chain.\\\\
    1152 An abort happens in any way, if the environment variable '''write_binary''' is not set to ''true'' and if moreover the job has been assigned an insufficient CPU time by '''mrun''' option {{{-t}}}.
     1152An abort happens in any way, if the activation string '''restart''' is not given with the {{{palmrun}}}-option {{{-a}}} and if the job has additionally been assigned an insufficient CPU time by '''palmrun''' option {{{-t}}}.
    11531153}}}
    11541154|----------------
     
    13491349The simulated time refers to the beginning of the initial run (t = 0), not to the beginning of the respective restart run. Restart runs can additionally be forced to be carried out in regular intervals using the run time parameter [#dt_restart dt_restart].\\\\
    13501350'''Note:'''\\
    1351 A successful operation of this parameter requires additional modifications in the '''mrun'''-call for the respective run (see [../runs Initialization and restart runs]).\\\\
     1351A successful operation of this parameter requires to set option {{{-a " ... restart ...}}} in the '''palmrun'''-call for the respective run (see also [../runs Initialization and restart runs]).\\\\
    13521352The choice of '''restart_time''' or '''dt_restart''' does not override the automatic start of restart runs in case that the job runs out of CPU time.\\\\
    13531353For [../examples/coupled coupled runs] this parameter must be equal in both parameter files [[../iofiles#PARIN|PARIN]] and [[../iofiles#PARIN_O|PARIN_O]].
     
    13821382{{{#!td
    13831383Number of processors along x-direction of the virtual processor net.\\\\
    1384 For parallel runs, the total number of processors to be used is given by the '''mrun'''-option -X. By default, PALM tries to generate a 2d processor net (domain decomposition along x and y), which is more or less square-shaped. If, for example, 16 processors are assigned (-X 16), a 4 * 4 processor net is generated ('''npex''' = 4, '''npey''' = 4). This choice is optimal for square total domains ([../inipar#nx nx] = [../inipar#ny ny]) because it minimizes the number of ghost points at the lateral boundarys of the subdomains. If nx and ny differ extremely, the processor net should be manually adjusted using adequate values for '''npex''' and '''npey'''.\\\\
     1384For parallel runs, the total number of processors to be used is given by the '''palmrun'''-option -X. By default, PALM tries to generate a 2d processor net (domain decomposition along x and y), which is more or less square-shaped. If, for example, 16 processors are assigned (-X 16), a 4 * 4 processor net is generated ('''npex''' = 4, '''npey''' = 4). This choice is optimal for square total domains ([../inipar#nx nx] = [../inipar#ny ny]) because it minimizes the number of ghost points at the lateral boundarys of the subdomains. If nx and ny differ extremely, the processor net should be manually adjusted using adequate values for '''npex''' and '''npey'''.\\\\
    13851385'''Important:'''\\
    1386 The value of '''npex''' * '''npey''' must exactly match the value assigned by the '''mrun'''-option -X. Otherwise the model run will abort with a corresponding error message.\\\\
     1386The value of '''npex''' * '''npey''' must exactly match the value assigned by the '''palmrun'''-option -X. Otherwise the model run will abort with a corresponding error message.\\\\
    13871387A specification of '''npex''' = 1 or '''npey''' = 1 overrides the default setting for the domain decomposition, i.e. it switches to a 1d decomposition, which might have a significant effect on the code performance. On machines with very slow communication network, the performance may improve, but usually it will degrade.
    13881388}}}