This repository contains the Python3 program SchedSim that can be used to simulate various energy efficiency scheduling strategies. It can digest traces from LoadLeveler and (better) SLURM.
- analysis - Contains plotting tools
- data/eex - This directory contains the hourly data generated for a paper
- data/test-trace - This directory contains a SLURM trace
- dev - Contains additional utilities
- e.g., to dump SLURM jobs as trace. New traces must be prepared using --prepareCSV.
- schedSim - The Core Python Code of SchedSim
- test - The binaries to execute Schedsim
The following command runs the example trace using the system configuration in data/test-trace $ ./run-test-trace-sim.sh
To see the available options, run: $ source dev/bash.src # This changes the PythonPath $ ./test/simulationSlurm.py -h
Note that when running a CSV, loading takes considerable amount of time. Additionally, the array of jobs can be stored using the Pickle format.
Conversion to Pickle can be done, for example, using: $ ./test/simulationSlurm.py --input data/test-trace/data.csv --convert data.p
New traces must be converted using: $ ./test/simulationSlurm.py --input slurm.csv --prepareCSV data.csv
Result of the simulation is stored in the file: output-stats.csv
This script generates statistics: $ ./analysis/statistics.R output-stats.csv
This script generates images of the utilization: $ ./analysis/utilization.R output-stats.csv
Interference of Billing and Scheduling Strategies for Energy and Cost Savings in Modern Data Centers https://doi.org/10.1016/j.suscom.2019.04.003
