In our Supercomputer Systems, you can execute a program in both a batch processing mode and an interactive processing mode suitable for developing or debugging a program.
For detailed procedures of each processing mode, see the following pages:
The System A require to create a static link to a specified library for executing programs. Thus, a program that is compiled in the System A cannot be executed in the System B and C. Conversely, a program that is compiled in the System B or C cannot also be executed in the System A.
When executing programs in the System A, you must describe an execution binary after the aprun command in either case of a non-parallelized program or a parallelization program. If you execute a program without the aprun command, the program runs on a gateway node, regardless of whichever processing mode you used for execution. Thus, please make sure that you execute a program using this command; otherwise, it disturbs other users.
When executing a program in a batch processing mode in our Supercomputer Systems, the amount of computing resources available for use varies depending on the Service Course you are using. For details, see Quota of Computing Resources.
On KUDPC, due to the prevention of disturbing other users' jobs, the total capacity of standard output files of jobs running at the same time is restricted to 8GB. If exceeding 8GB, the jobs running at the same time are forced to be halted.
If your program will output a lot of data, please make the program redirect output to files as follows.
Ex) on System A:
aprun (*snip*) ./a.out > output.log
A job whose time limit is set to within one hour by -W option can get compute nodes more preferentially than that whose time limit is set to over one hour.
If a job's duration time is expected to be within one hour but it's time limit is set to over one hour, the job may be executed more quickly by setting the time limit to within one hour.
The default time limit of job is one hour, so a job with no -W option can get compute nodes preferentially.