Glossary

Last updated on 2026-10-27 | Edit this page

node: One computer in the cluster. --nodes is how many you ask for.
task: One MPI rank. --ntasks is how many ranks the job may launch. srun starts them. sbatch only reserves the allocation.
thread: One OpenMP thread inside a rank. The hybrid scripts set OMP_NUM_THREADS from SLURM_CPUS_PER_TASK.
core: A hardware processor. --cpus-per-task is how many cores Slurm reserves for the threads of one task.
AXPY: The update y = a * x + y, applied element by element. There is no reduction.
dot product: The sum of x[i] * y[i]. The hybrid program reduces that sum with OpenMP and with MPI.
reduction: Combining partial sums into one sum. In this lesson the OpenMP clause is reduction(+:) and the MPI call is MPI_Allreduce with MPI_SUM, the all-ranks form of MPI_Reduce.
checksum: The sum of every entry of C after matrix multiplication. The program prints it as Matrix checksum: with %.17g. Same n, same line.
InfiniBand: The network that connects Snellius compute nodes. The course machines use Ethernet, so a job that crosses nodes scales more weakly there.