We introduce the prototype of MDDash -- a web application to run and publish molecular dynamics calculations. This is not the first attempt to do so, but our emphasis diverges from the usual convenience web portals. We focus on efficient use of computational resources, setup of the run parameters (configuration of CPU cores and GPU use etc.) is tuned automatically. Calculations are managed in a reproducible way, the whole sequence (captured in a Jupyter notebook) becomes part of a provenance record. The dashboard also integrates with InvenioRDM repository, allowing the user to create draft repository records from the calculation results directly. Finally, the architecture is extensible, users are expected to contribute different simulation and analysis workflows. After successfull authentication, the user is presented with a wizard-style interface with five main stages: 1) preparation, where the simulation is set up (lightweight calculations, typically up to NPT/NVT equilibration), within a Jupyter notebook which records all the steps; 2) tuning, where short runs (few picoseconds) of the prepared simulation are tried with different number of MPI processes, OMP threads, and assignment of calculations (PME, PP) to GPUs to evaluate performance and to choose the best performance/resources tradeoff; 3) production simulation, run in the tuned setup; 4) analysis including visualization; 5) publishing to repository. Technically, the dashboard is implemented on top of Jupyter Hub in Kubernetes, which provides authentication (via OAuth2) and elementary isolation of the users. Extensible parts of the workflow (preparation and analysis) are Jupyter notebooks; we provide initial ones, users are encouraged to contribute their flavors. The publication step involves calling our Gromacs Metadump tool (https://doi.org/10.1186/s13321-025-01082-5) which extracts extensive metadata set from the simulation files automatically. Currently, the dashboard supports Gromacs simulations, however, the architecture allows planned extensions to other MD engines. We provide a testing instance at https://mddash.dyn.cloud.e-infra.cz/ (accessible on request), all the code including complete Helm charts for users' deployment is available at https://github.com/CERIT-SC/mddash/. MDDash is also being integrated among the supported target environments of EOSC Data Commons services; among others the integration will allow searching for MD datasets in the repositories via EDC AI-enabled search, and analyzing or reproducing these results in MDDash smoothly.
Krása et al. (Thu,) studied this question.