DAG Scheduling Simulator

This python script simulates the allocation of the nodes of an OpenVX DAG on any number of CPUs and GPUs using G-FL(0), EDD(1) or HEFT(2)/XEFT(4) scheduling. A custom version of G-FL is also available(3), which uses G-FL normally for scheduling, but uses the HEFT mapping instead of the default approach (VisionWork style).

Requirements

In order to run these scripts the following libraries are required:

matplotlib
scipy

Usage

sim.py FILENAME[csv] SCHEDULER(0,1,2,3,4) FRAMES(n) PIPELINE(0,1)

Examples:

Simulates using G-FL , 3 frames and no pipeline, loading from "graph_file.csv" (Default settings)

sim.py

Simulates using EDD and 2 frames, loading from "graph_file.csv", using the pipeline

sim.py 1 2 1

Simulates using HEFT, 10 frames and pipeline, loading from "graph.csv"

sim.py graph.csv 2 10 1

Auto simulation with generated graphs

Simulates a number of generated graphs (using generator.py and sim.py), output in "output.csv"

auto_sim.py

Re-simulate with past simulations

Exctracts additional data (now configured for memory footprint), has to be run in the same folder as the *.pkl files generated by sim.py, outputs to "result.csv", use "merge.py" to merge with "output.csv"

re_sim.py