Glossary

Terms used throughout this documentation, explained for users who are new to HPC.

Term	Meaning
Chunk	A part of the resources requested by a job that is placed on a single node, e.g. select=2:ncpus=8 requests two chunks. See Resource Allocation.
Checkpoint	A file in which a program saves its state to continue later. See Long Jobs.
Compute node	A server on which jobs run. The cluster has 17 of them, see Compute Nodes.
Container	A file containing a program with all its dependencies, run with Singularity.
Core	A physical processing unit of a processor. Each core runs two hardware threads.
CPU (ncpus)	In PBS, one logical CPU, i.e. one hardware thread.
Environment	An isolated set of installed packages, e.g. a Conda environment or a Python virtual environment.
Exit status	A number returned by a finished job, 0 means success. See Exit Status.
Fair-share	Scheduling policy that gives higher priority to users who have used less of the cluster recently. See Job Scheduling.
Home directory	Your personal directory ~ for code and configuration. See Storage.
Interactive job	A job that gives you a terminal on a compute node. See Interactive Jobs.
Job	A computation submitted to the scheduler with qsub.
Job array	Many similar jobs submitted at once. See Job Arrays.
Job script	A Bash script with #PBS directives describing a job.
Login node	The server login.hpc.imtm.cz you connect to. Used for editing, data transfer and submitting jobs, not for computations.
Module	Software prepared by administrators, loaded with module load. See Lmod.
MPI	A standard for programs that run on several nodes at once. See Parallel Jobs.
OpenMP	A standard for programs that use several threads on one node.
PBS / OpenPBS	The scheduler of the cluster, which decides when and where jobs run.
Project directory	Storage shared by the members of a project in /project. See Storage.
Queue	A list of jobs waiting to start. Each queue may have its own limits.
Scratch	Fast temporary storage for a running job, deleted when it ends. See Scratch Space.
Thread	A sequence of instructions running on one logical CPU. A program with 8 threads can use 8 CPUs.
Walltime	Maximum real time a job may run. The job is killed when it is exceeded.