Skip to content

Latest commit

 

History

History

pmembench

README.md

The pmembench benchmark (Duration ~45-60mins)

Workflow description

We run the vanilla PMDK, SPP and SafePM variants of the phoenix benchmarks using the already built images.

The run_variants.sh script executes the pmembench benchmark for each variant and places the results in the created results directory and further subdirectories with names that showcase the benchmark name. The respective files inside the subdirectories define the variant. Note that the repeats parameter here is taken from the benchmark configuration; an example can be found here. By default, the amount of repeats is set to 3 for the PM indices (pmembench_map.cfg) and 10 for the PM TX (pmembench_tx.cfg) operations.

Note: If the machine you are using has multiple NUMA nodes, it is mandatory that you specify the node that the mounted PM is attached. This can be done via the NUMA_NODE environment variable. Otherwise, you can ignore this variable.

To run the benchmark, use:

$ cd PROJECT_ROOT/benchmarks/phoenix
$ NUMA_NODE=0 BENCHMARK_PM_PATH=<path to PM> ./run_variants.sh

For more information about the images and containers, see here.

Plots

To generate the plots based on the results, simply run:

$ cd PROJECT_ROOT/plot_utils
$ ./import_pmembench.sh <path-to-pmembench-results>
$ python3 generate_plots.py plot_config_pmembench_tx.yml
$ python3 generate_plots.py plot_config_maps.yml

The first import_pmembench.sh script mainly copies the results of the original folder and formats it so that they can be used by the plotting scripts. The generate_plots script will generate a bunch of plots for the pmembench results which are placed in PROJECT_ROOT/plot_utils/plots directory. For each benchmark, a new subdirectory is created. The executions above will also print a summary of the analyzed result in the stdout.

For more information about the plotting scripts, see here.

File description

  • run_safepm.sh: Script that runs inside the newly-built container for SafePM and modifies the appropriate pmembench benchmark files to set the correct paths for the PM pools and executes the pmembench benchmark.
  • [run_spp.sh](./run_spp.sh): Script that runs inside the newly-built container for SPP and modifies the appropriate pmembenchbenchmark files to set the correct paths for the PM pools and executes thepmembench` benchmark.
  • run_vanilla.sh: Script that runs inside the newly-built container for vanilla PMDK and modifies the appropriate pmembench benchmark files to set the correct paths for the PM pools and executes the pmembench benchmark.
  • run_variants.sh: Script that instruments the execution of the pmembench benchmark for all the variants and configurations.

Note: We further provide scripts to run pmembench in additional environments:

  • *_O1.sh: These scripts (with the suffix _O1) run pmembench in the same environment but set the optimization level to O1 using the respective O1 images, produced by the scripts here.
  • *_mock_*.sh: These scripts run pmembench in the SPP environment where we rule out the SPP pointer instrumentation by forcing the SPP injected function to immediately return without performing any action. They use the respective mock images, produced by the scripts here.