Skip to content

Parallelize Benchmark execution #843

Description

@bouthilx

The benchmark generally contains large amount of experiments. Even if the tasks are fairly short the execution of the benchmark overall takes a very long time unless we parallelize it. It is possible to execute the same benchmark multiple times in parallel but it is cumbersome. We should support parallelizing the execution of the experiments while executing benchmark.process, by reusing an Executor object.

The trickiest part is that some assessments (ex ParallelAssessment) can also parallelize the execution of trials. These assessments' executor should not be starved by running too many experiments in parallel, thus we need to account for the ressources needed by the assessment when dispatching the experiment.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementImproves a feature or non-functional aspects (e.g., optimization, prettify, technical debt)

    Type

    No type

    Projects

    No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions