HPC cluster is available 24/7, and since simultaneous use of the cluster by several users is possible, virtual job queues have been introduced. It means – if resources needed for a job (processor cores or memory) exceeds the amount of available HPC cluster resources, the job will be put in a queue. The queue time depends on the number of jobs waiting in the queue, as well as the amount of resources claimed by the user.
In order to balance the HPC cluster load between users, the HPC Center’s admins hold the right to change the amount of HPC cluster resources (for example, processor cores) available to a single user or the number of jobs to be executed simultaneously.
The user must avoid activities that might interfere with the work of other users (queue bypassing, exceeding the reserved amount of resources, etc.).
If possible, avoid execution of jobs locally on the login node (ui-1.hpc.rtu.lv). Use it only to submit a job to a queue, to compile programs, and test short jobs.
Try to indicate the necessary walltime so that other users can forecast their queue time.
If the job entails intensive disk operation (I/O), it is advised to use a local SSD disc instead of the network directory (/home/*). The local disc is mounted to the directory /scratch. When the job is completed, the files created in this directory should be removed manually.
Evaluate the performance of parallel (MPI) jobs by using various number of nodes / cores to select the most suitable one for your job. Larger number of cores does not always ensure faster execution of a job.