Chameleon benchmarks

The scripts located here are used to analyze Chameleon performances in a systematic way. The workflow is as follows:

1 Gitlab-ci

We use a schedule job that will be triggered with a chosen frequency, the schedule job frequency can be tuned in the CI/CD part in the web interface

2 Guix

Guix is responsible for building an isolated and reproducible environment to build and execute Chameleon, Jube and some python scripts. We can use it on PlaFRIM because Guix and Guix-HPC are installed.

3 Jube

Jube is used to drive the execution with different parameter spaces and to parse the results in csv files. Appart from the execution parameters such as the problem sizes the number of ressources used and so on we also save in the database the commit date of chameleon and the commit ids of chameleon and guix channels to properly identify the software versions.

4 Elasticsearch

Elasticsearch is the database framework. The server is https://elasticsearch.bordeaux.inria.fr. It is only accessible from Inria’s networks for now.

5 Kibana

Kibana is a web server to visualize the performances on graphs. It looks for data imported in the elasticsearch database. We want to be able to analyze the performances for each commit for which the scheduled job has been performed and to monitor some performances in the course of time/commits. Kibana server is hosted here https://kibana.bordeaux.inria.fr. It is only accessible from Inria’s networks for now.