Skip to content

old online workloads

helq edited this page Jan 30, 2024 · 1 revision

1- Checkout the SWM workloads and build them as follows. Note that on some machines, this would need swapping a newer gcc version 7.1 or above. With older gcc versions, you might get the error 'nullptr not recognized'.

../configure --prefix=/home/mubarak/codes-online/workloads/swm/install

make

make install

Note that you will need boost for building SWM workloads.

2- Checkout argobots and build as follows:

git clone https://github.com/pmodels/argobots.git

./autogen.sh

../configure --prefix=/home/mubarak/codes-online/argobots-install

3- Checkout the codes branch 'workloads' and configure as follows:

../configure --with-online=/home/mubarak/codes-online/workloads/swm --with-boost=/soft/libraries/boost/1.57.0/intel-mvapich2/ PKG_CONFIG_PATH=/home/mubarak/codes-online/argobots-install/lib/pkgconfig/:/home/mubarak/codes-online/ROSS/install/lib/pkgconfig/:/home/mubarak/codes-online/workloads/swm/install/lib/pkgconfig/ --prefix=/home/mubarak/codes-online/codes/install CC=mpicc CXX=mpicxx CFLAGS=-g CXXFLAGS=-g

4- Run the simulation as follows:

./src/network-workloads/model-net-mpi-replay --sync=1 --workload_type=online --extramem=1310720 --workload_conf_file=workloads.conf --lp-io-dir=lammps-test-1 --lp-io-use-suffix=1 --alloc_file=allocation.conf -- ../src/network-workloads/conf/dragonfly-custom/modelnet-test-dragonfly-theta.conf

In workloads file, indicate 'lammps' or 'nekbone'.

5- Scaling the lammps and nekbone workloads:

In workloads/swm/src/lammps/lammps_workload.json, increase the num_x_replicas, num_y_replicas and num_z_replicas for larger ranks (currently set to 3 for 1,024 ranks).

In workloads/swm/src/nekbone/workload.json, increase the Rx, Ry, Rz, Ex, Ey and Ez parameters and make sure RxRyRz is equal to number of simulated ranks.

Clone this wiki locally