|
|
|
When ADF has been installed you can run it by supplying appropriate input and starting the binary executable ("adf.exe" in $ADFBIN). For a parallel run a few things have to be arranged first, such as the `export' of some environment variables to the parallel processes. All this is arranged by a run script "adf", located in $ADFBIN. You can use this run script also for the serial version. For other programs in the package, there are similar run scripts ("band", "dirac", and so on). We recommend to always apply the script, rather than the binary executable directly. Running the program using the run script involves the following steps:
During the run you may inspect the logfile, to see how far the program has proceeded, or whether you should interrupt the calculation.
In the above scheme adf is the name of the run script that invokes the adf.exe program executable. During the installation the script has been put in the same directory where the program executables are generated: $ADFBIN. You may have moved it to another place, or renamed it. We recommend that you adjust your $PATH variable so that you can omit $ADFBIN from the execution command.
To run another program from the ADF suite, just use the appropriate program run script.
The input for the program is read from standard input, and output is written to standard output, redirected in the example above to in and out, respectively.
The part between square brackets is optional and is only meaningful for a
parallel program version. The -n flag specifies the number of parallel
processes (nproc) to use. If omitted the default applies, which is the value of
the environment variable $NSCM, if such variable exist, otherwise it is defined
by installation parameters (in the $ADFHOME/settings file, see the Installation
Manual).
Irrespective of the -n value, the $NSCM variable value and any installation parameters, two items define an absolute upper bound for the number of parallel processes that can be used. The first is given (PVM) by the so-called virtual parallel machine, i.e. by the number of hosts that constitute the virtual parallel machine. See the PVM manuals. The second item is the maximum number of parallel processes that you allow to run on each particular host that is part of the virtual parallel machine. This information must be stored in a file nodeinfo, located in the $ADFBIN directory. For any host not mentioned in this file, the maximum number of parallel processes on that host is one (1). Note that all hosts to be included must be known to PVM as being part of the virtual parallel machine, otherwise no process can be spawned on that host at all.
Two examples may illustrate this. First: on an IBM SP you should run one process per node. Therefore, no file nodeinfo is needed. Second: on an SGI Power Challenge with 4 CPUs, you should create a file $ADFBIN/nodeinfo, with one record: the name of the host (as it is known to PVM, consult your PVM manual, or see our Installation manual) followed by one or more spaces and then the number 4.
The program run scripts have, in fact, more flags and arguments, for special usage. You can get a survey by typing
$ADFBIN/adf -h
A note for those who are used to the ADF 2.3 release, where the run script start was used. The start script still exists (in $ADBIN) and can be used as before (more or less), but this is not necessary anymore. All program run scripts are links to the start script.
The ADF program may generate several output / result files, along with the standard output file. The most important one is TAPE21, the general result file. TAPE21 contains relevant information about the outcome of the calculation. You may want to store this file somewhere under an appropriate name, for future usage. The meaning of any other files that are produced are explained later in this User's Guide.
We will start now with a discussion of the input file for ADF.