This repository provides the time-energy data collected and analyzed in our research paper entitled "The Time and Energy Efficiency of Modern Multicore Systems" (PARCO 2019). Please read our Data-in-Brief article for a detailed description of the data.
To generate model data in data/model
folder, go to scripts folder and run:
$ ./run-homogeneous.sh
$ ./run-heterogeneous.sh
You need bash
and Python 3
to run these scripts.
The model's equations are implemented in model.py.
The script model_homogeneous.py uses the equations corresponding to homogeneous multicore systems implemented in model.py. It takes as paramters the number of cores, the active power fraction of the system (APF), the idle power of the system (in Watts), application name (for display) and the measured data for the given application on the given homogeneous system.
Usage: model_homogeneous.py n apf psys app measurements_file
n - total number of cores
apf - APF
psys - system power [W]
app - application name
measurements_file - time-energy measuremnets on all cores
We provide a set of parameters for the AMD, ARM, Xeon, i7 and Pi3 homogeneous systems [1, 2, 3] in conf-amd.sh, conf-arm.sh, conf-xeon.sh, conf-i7.sh, conf-pi3.sh, respectively.
The script model_heterogeneous.py uses the equations corresponding to heterogeneous multicore systems implemented in model.py. It takes as paramters the total number of cores, the number of little and big cores, the active power fraction (APF) of little and big cores, the idle power of the system (in Watts), application name (for display) and the measured data for the given application on little cores, big cores, all cores using static scheduling and all cores using dynamic scheduling.
Usage: model_heterogeneous.py n nl nb apf_l apf_b app file_l file_b file_static file_dynamic [-l2b]
n - total number of cores
nl - number of little cores
nb - number of big cores
apf_l - APF of little core
apf_b - APF of big core
p_sys - system (idle) power
app - application name
file_l - performance measuremnets on little cores only
file_b - performance measuremnets on big cores only
file_static - performance measuremnets on all cores using static OpenMP scheduling
file_dynamic - performance measuremnets on all cores using dynamic OpenMP scheduling
We provide a set of parameters for the XU3 and TX2 heterogeneous systems [1, 2, 3] in conf-xu3.sh, conf-tx2.sh, respectively.
-
[1] D. Loghin and Y.M. Teo, "The Energy Efficiency of Modern Multicore Systems", in Proc. of the 47th International Conference on Parallel Processing Companion https://doi.org/10.1145/3229710.3229714
-
[2] D. Loghin and Y.M. Teo, "The Time and Energy Efficiency of Modern Multicore Systems", in "PARCO" https://doi.org/10.1016/j.parco.2019.04.009
-
[3] D. Loghin and Y.M. Teo, "Time-energy Measured Data on Modern Multicore Systems Running Shared-memory Applications", in "Data in Brief" https://doi.org/10.1016/j.dib.2019.104670).
This repository is licensed under Creative Commons Attribution 4.0 license (http://creativecommons.org/licenses/by/4.0/)
If you are using these data in your research, please cite one of our papers.