Usage guidelines

<html> Fast local SSD disk scratch available with gpu nodes (gpu[001-021]) (mounted at /tmp). </html>

<html> With exclusive jobs on large SMP nodes, temporary files could just be placed into /dev/shm/<path/to/your/folder> (in-memory tmpfs). </html>

<html> Please especially avoid copying around TB large rwf and chk files from Gaussian runs. Please remove temporary files from scratch at the end of job runs. </html>

<html> Please put volatile job data at /scratch/<account-name> or in the working directory of your job and delete temporary data at the end of job runs. </html>

<html> Please carefully calculate the memory limit for your jobs, so that optimal scheduling of your jobs can happen. </html> * <html> There is no virtual memory limit applied to your jobs, only real memory usage is accounted. This is subject to eventual change in future. </html> * <html> Defining virtual memory limits: The standard memory allocation parameter –mem defines the total amount of memory available to all processes on one node. For advanced usage, it's also possible to define the amount of memory allocated per process reps. cpu (c. f. man 1 sbatch or man 1 srun). </html> * <html> In general, hyper-threading should not be used. Please do not set the parameter –threads-per-core or –cpus-per-task, unless you really know what you do. </html>

date of revision: 06-24-2019 © kraus