High-Performance Computing, Beyond: Cheatsheets for Queuing System Quick Reference

Key Points

Compiling on a cluster	Compilers turn instuctions in programming languages to machine code that the computer can execute
Introduction to Parallelism	Parallelism can speed up the execution of your program The structure of your code, and scheduling considerations, will influence how you parallelize your computational work
Compiling MPI code on a cluster	mpicc compiles MPI code written in C mpif90 compiles MPI code written in Fortran (90) mpirun is used to run MPI programs, specifying the number of processors with the -np flag The rank of an MPI process does not determine which process starts or finishes first
Compiling OpenMP code on a cluster	`gcc` with the `-fopenmp` flag compiles OpenMP code written in C `gfortran` with the `-fopenmp` flag compiles OpenMP code written in Fortran The program generated is run as usual, but the number of processors is controlled with the `OMP_NUM_THREADS` environment variable
Benchmarking and Optimization	Deciding how many processors to run on is an iterative task Speedup and efficiency are two measures that can help us figure out how many processors to run on

Cluster: a collection of computers configured to enable collaboration on a common task by means of purposefully configured hardware (e.g., networking) and software (e.g. workload management).