We are experimenting scheduling strategies to allocate resources to workloads with widely different resource consumption between tasks. A side product of the work storing in this repository is a paper accepted to the WORKS workshop 2021 (to be updated later).
You can find all synthetic and real datasets under the resources_data directory.
Running make {dataset} where dataset is in {colmena, hypersweep, exponential, normal_large, normal_small, bimodal, trimodal, uniform_large, uniform_small} will generate respective results in resources_analysis/{dataset}/.
4 levels of information: nothing -> number of classes -> tag for each task -> everything