Skip to content

Latest commit

 

History

History
57 lines (43 loc) · 2.69 KB

running.md

File metadata and controls

57 lines (43 loc) · 2.69 KB
Raw

Running Grappa programs

Grappa jobs are run like any other MPI job. This process varies depending on how your cluster is configured and what version of MPI you use. Here are some examples.

Note that running MPI jobs like Grappa is very system-dependent. Talk to your system administrator to learn the correct commands for running process-per-core jobs like Grappa apps.

Slurm with srun

Most of the systems we run on allow MPI jobs to be launched with srun directly:

srun --nodes=2 --ntasks-per-node=2 -- applications/demos/hello_world/hello_world.exe

Slurm with salloc/mpirun

Some MPI installations aren't configured to work with srun; instead, they require first getting an "allocation" from the scheduler and then calling "mpirun" or "mpiexec" inside that allocation:

salloc --nodes=2 --ntasks-per-node=2 -- mpirun applications/demos/hello_world/hello_world.exe

Mpirun with hostfile

Some MPI installations aren't configured to work with srun; instead, they require first getting an "allocation" from the scheduler and then calling "mpirun" or "mpiexec" inside that allocation. On these systems you have to make sure the task count and host count are appropriate to get you whatever number of tasks per node you expect.

mpirun -n 4 -hosts=<list of 2 hosts> applications/demos/hello_world/hello_world.exe

Legacy information: using Slurm and grappa_srun

The Grappa team built a tool with some shortcuts for running on machines they commonly use during the development process. It's not necessary to use this tool to run Grappa jobs---any MPI job launcher will work. These days, we usually just use srun directly.

But for posterity, here's more info: the Ruby script bin/grappa_srun can be used to automatically manage the pesky srun flags that are needed to run Grappa programs correctly on a number of machines that we've tested on. Try running bin/grappa_srun --help for more detailed usage information.

Running an application:

# in build/Make+Release: #
# build the desired executable #
make -j graph_new.exe
# launch a Slurm job to run it: #
bin/grappa_srun --nnode=2 --ppn=2 -- graph_new.exe --scale=26 --bench=bfs --nbfs=8 --num_starting_workers=512

Running a Grappa test:

# in your configured directory, (e.g. build/Make+Release) #
# (builds and runs the test) #
make -j check-New_delegate_tests
# or in two steps: #
make -j New_delegate_tests.test
bin/grappa_srun --nnode=2 --ppn=2 --test=New_delegate_tests