Adding a New Machine

The following steps can be used to add support for a new system:

  1. Create a machine configuration file (e.g. lyra.ini, placed in $OLCF_HARNESS_DIR/configs/lyra.ini).

  2. Create the location of the repository that will hold all applications and tests for that particular machine

    • This can be in a remote Git repository, or a directory on the file system (which is then used as Path_to_tests in the input file)

Creating a new machine configuration file is discussed below. The creation of new tests is discussed in Adding a New Test.

Create a Basic Machine Configuration

An example input file with detailed explanation is contained below. All settings in the machine.ini file are used to set environment variables that can be used throughout the test. For example, gpus_per_node is used to set RGT_GPUS_PER_NODE, which could be used in a job script. These settings do not override existing environment variables. If RGT_SCHEDULER_TYPE is set by the user, then the machine.ini file will not override it.

[MachineDetails]
# Required variables :
machine_name = frontier
# options: linux_x86_64 or ibm_power9
machine_type = linux_x86_64
# options: slurm, pbs, lsf
scheduler_type = slurm
# options: srun, aprun, jsrun, poe
joblauncher_type = srun

# Optional: specify some details about the machine
# Each of these becomes an environment variable,
# ie **RGT_CPUS_PER_NODE**, that can be used to template tests
# These can also be set in SiteCustom, but the harness does attempt to read these
# The harness just doesn't act on these yet
node_count = 9408
cpus_per_node = 56
sockets_per_node = 1
gpus_per_node = 8

# This section is only required when ``--mode checkout`` is used
[RepoDetails]
# Git is currently the only supported type of remote repository
type_of_repository = git
# Branch/tag to clone from remote repo -- we recommend NOT setting this unless needed
#git_reps_branch = master
# Protocol to use when cloning
git_data_transfer_protocol = ssh
# Sub-project/group in the Git repository
git_server_application_parent_dir = your-project/applications
# Name of the machine to look for in Git.
# OTH looks for:
#   $git_ssh_server_url:$git_server_application_parent_dir/$git_machine_name/<application>
git_machine_name = frontier
# Git URLs:
git_ssh_server_url = git@github.com
git_https_server_url = https://www.github.com

# This section provides defaults for testshot variables
[TestshotDefaults]
# The path used for building the application and scratch space used for running
path_to_sspace = /default/path/to/scratch/space
# A string that can be used to identify tests run for a specific purpose (ie: 'summit_tshot_cuda11')
system_log_tag = frontier_test
# Default queue/partition to submit jobs to
batch_queue = batch
# Overridden if project_id is set in the test's input file
project_id = <default account for scheduler>
# Any required flags to job scheduler (ie, ``-M clustername`` in Slurm)
submit_args =

# This optional section is where you can define custom environment variables you might use
[SiteCustom]
# An example usage could be providing a variable to tell your code what the name of the MPI module is
my_mpi_module_name = openmpi/4.1.6

Note

These variables can all be overridden by setting RGT_<VARIABLE_NAME> in the environment prior to launching. However, that usage of batch_queue/RGT_BATCH_QUEUE and project_id/RGT_PROJECT_ID may be incorrect. For handling these variables, please see Run-time environment parameters.