A Python 3.6+ implementation of a heuristic DAG scheduling approach from
H. Arabnejad and J. G. Barbosa, "List Scheduling Algorithm for Heterogeneous Systems by an Optimistic Cost Table," in IEEE Transactions on Parallel and Distributed Systems, vol. 25, no. 3, pp. 682-694, March 2014. doi: 10.1109/TPDS.2013.57
Setting up a virtual environment first is recommended
After that, all necessary dependencies can be installed with pip install -r requirements.txt
Otherwise, the main dependencies are:
- Python 3.6+ (uses literal string interpolation)
- Matplotlib
- Numpy
- Networkx
- Pytest (development dependency only)
- nb_conda (only if using a conda environment & want jupyter notebook to use the right python version)
Finally, if you want it available as a local package for availability elsewhere on your system, it can be installed with pip install .
Basic usage is given by python -m peft.peft -h
usage: peft.py [-h] [-d DAG_FILE] [-p PE_CONNECTIVITY_FILE]
[-t TASK_EXECUTION_FILE]
[-l {DEBUG,INFO,WARNING,ERROR,CRITICAL}] [--showDAG]
[--showGantt]
A tool for finding PEFT schedules for given DAG task graphs
optional arguments:
-h, --help show this help message and exit
-d DAG_FILE, --dag_file DAG_FILE
File containing input DAG to be scheduled. Uses
default 10 node dag from Arabnejad 2014 if none given.
-p PE_CONNECTIVITY_FILE, --pe_connectivity_file PE_CONNECTIVITY_FILE
File containing connectivity/bandwidth information
about PEs. Uses a default 3x3 matrix from Arabnejad
2014 if none given.
-t TASK_EXECUTION_FILE, --task_execution_file TASK_EXECUTION_FILE
File containing execution times of each task on each
particular PE. Uses a default 10x3 matrix from
Arabnejad 2014 if none given.
-l {DEBUG,INFO,WARNING,ERROR,CRITICAL}, --loglevel {DEBUG,INFO,WARNING,ERROR,CRITICAL}
The log level to be used in this module. Default: INFO
--showDAG Switch used to enable display of the incoming task DAG
--showGantt Switch used to enable display of the final scheduled
Gantt chart
If you don't have any particular DAG that needs scheduling, the canonical example schedule from Topcuoglu et al. can be generated by passing in no args
python -m peft.peft
With a generated Gantt chart available using
python -m peft.peft --showGantt
If Pytest is installed, tests can be executed simply by running pytest from the repository root directory