JUPITER Benchmark Suite: Arbor

This benchmark is part of the JUPITER Benchmark Suite. See the repository of the suite for some general remarks.

This repository contains the Arbor benchmark, created by CSCS and JSC. DESCRIPTION.md contains details for compilation, execution, and evaluation. The benchmark was developed together with Arbor contributors, especially CSCS.

Sources are included as submodules.

Quickstart using JUBE

# Run the benchmark using JUBE
jube run benchmark/jube/default.yaml --tag baseline
jube result -a benchmark/jube/run --id XYZ

This will perform a full build and run the benchmark afterwards. On JUWELS Booster jube run default.yml --tag baseline produces the following output

cells	...	ranks	...	t_init	t_run	n_spike
1536000	...	32	...	720.519	727.294	1549882

Some fields omitted for brevity.

Potentially, a platform configuration file (platform.xml) for JUBE is needed.

For a more detailed description, see DESCRIPTION.md.

Without JUBE

You will need to build the required software modules in src/arbor and src/nsuite such that the busying benchmark in nsuite links against libarbor.a from src/arbor. Refer to the JUBE script on how this is done. Then, generate the input files for the benchmark using benchmark/gen-inputs.py --default and run busyring passing one of the resulting JSON files.

We provide static input files in benchmark/inputs for 50 PF and 1 EF, as well as the baseline. See the JUBE script benchmark/jube/default.yaml and DESCRIPTION.md for more details.