A Python 3.6+ implementation of a heuristic DAG scheduling approach from
H. Topcuoglu, S. Hariri and Min-You Wu, "Performance-effective and low-complexity task scheduling for heterogeneous computing," in IEEE Transactions on Parallel and Distributed Systems, vol. 13, no. 3, pp. 260-274, March 2002.
Includes alternative heuristics that were leveraged as a part of this work:
J. Mack, S. E. Arda, U. Y. Ogras and A. Akoglu, "Performant, Multi-Objective Scheduling of Highly Interleaved Task Graphs on Heterogeneous System on Chip Devices," in IEEE Transactions on Parallel and Distributed Systems, vol. 33, no. 9, pp. 2148-2162, 1 Sept. 2022, doi: 10.1109/TPDS.2021.3135876.
Setting up a virtual environment first is recommended
After that, all necessary dependencies can be installed with pip install -r requirements.txt
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 heft.heft -h
usage: heft.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 HEFT 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 Topcuoglu 2002 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 Topcuoglu
2002 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
Topcuoglu 2002 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 heft.heft
With a generated Gantt chart available using
python -m heft.heft --showGantt
Example usage as an external library is given by this notebook
If Pytest is installed, tests can be executed simply by running pytest from the repository root directory