Evaluation of Datacenter Scheduler Programming Abstractions

## Evaluation of Datacenter Scheduler Programming Abstractions

This is an archive of the Github repository containing the evaluation code and experiments written in Kotlin for the research work that investigates the performance impact of various datacenter scheduler programming abstractions.

### Prerequisites
- An x86 machine with at least 16GB RAM.
- Any desktop Linux distribution with build tools installed. For example, `build-essential` package on Ubuntu.
- Java 17 or greater.
- Python 3.11 (We recommend conda, virtualenv, or a similar environment).

### Reproducibility instructions

- Extract the repository from the zip file.
- Run `./download-traces.sh`. This will download all required traces from Zenodo and copy them to the relevant directories for experiments.
- Setup a python 3.11 virtual environment. Install dependencies using `pip install -r plot/script/requirements.txt`.
- Use `./run-experiments.sh` to run the simulations and plot the results.
- Figure 7 in the paper is `plot/output/migrations-results-packing-azure.pdf`. Figure 8 is `plot/output/migrations-results-totaltime-azure.pdf`. Figure 10 is `plot/output/metadata-results-ibm.pdf`.