FASRC specific configuration for Reframe

These are configs for different clusters run by FASRC.

• cannon: Configs for cannon.
• fasse: Configs for fasse.
• test: Configs for the test cluster.

Checks currently are copied from the cscs-checks folder in reframe and tuned to the FASRC environment.

These are benchmark tests that are broken down by category. Here is a short description of each test:

• alloc_speed: Tests speed of memory allocation. Originally from CSCS
• dgemm: Runs dgemm code to get a measure of FLOps. Originally from CSCS
• latency: Measures latency to L1, L2, L3 cache. Originally from CSCS
• stream: Runs STREAM test for measuring memory bandwidth. Originally from CSCS
• strided_bandwidth: Runs bandwidth test with various stride sizes. Originally from CSCS

• dgemm: Runs dgemm code to get a measure of FLOps. Originally from CSCS
• gpu_burn: Burns in GPU and gives a report of GFLOps for one GPU and its temperature. Originally from CSCS
• kernel_latency: Tests latency in loading the NVIDIA kernel. Originally from CSCS
• memory_bandwidth: Tests GPU memory bandwidth. Originally from CSCS
• pointer_chase: Runs a linked list search to test memory latencies. Originally from CSCS
• shmem: Tests shared memory bandwidth. Originally from CSCS

• fft: Test runs FFTW. Originally from CSCS
• halo_exchange: Simulates halo cell (aka ghost or boundary zone) exchange to test MPI communications. Originally from CSCS
• hpcg_benchmark: Runs the HPCG benchmark for gnu and MKL. Originally from CSCS
• osu: Various MPI benchmarks from OSU. Originally from CSCS

These are codes from FASRC User_codes repo. We try to test the main software used on the cluster.

• AI
  • PyTorch: pulls PyTorch Singularity image and runs check_gpu.py inside the container
  • TensorFlow: pulls TenforFlow Singularity image and runs tf_test_multi_gpu.py inside the container
• Languages
  • Cpp: runs dot_prod.cpp
  • Python
    • Example1/monte_carlo_pi.py runs mc_pi.py
    • Example3/mamba_env.py runs build_env.sh to build a mamba environment and numpy_pandas_ex.py to use python packages installed in the mamba environment
  • R: runs count_down.R
• Parallel computing
  • Matlab: runs parallel_monte_carlo.m

The reframe tests that run the codes listed above were modified tests from Reframe tutorial and from University of Southern California uschpc reframe repo. These tests can be run either from a login or a compute node as ReFrame runs these tests by submitting jobs to the Slurm scheduler.

These tests are to verify various aspects of the system. Here is a short description of each test.

• io: Test runs IOR on several filesystems. Originally from CSCS
• slurm: Various slurm sanity checks. Originally from CSCS

https://github.com/eth-cscs/reframe

https://reframe-hpc.readthedocs.io/en/stable/index.html