Do we need model runtime restrictions for the modeler, if the model is supposed to be uploaded?

[llavall]

I would like to include in the rules list on "what to take care of if you create a model in order to upload it to the model repository" that the model has to be designed and described in a way that it is fully functional, but has as less runtime as possible. But the user must be able to understand which settings are needed to calculate the original model results. The background is, that it is wanted that the testing strategy includes that all models are run before and after a build is published, in order to find problems. If we take Tasjas model as example, it might take 7 days to fully run it. Her minimal version provided did 3 instead 1000 loops and the model needs only 3 minutes. Maybe it would make sense to include 2 simulations in the model original vs. minimal for testing.

Best wishes Lars

[mfilter]

Very good suggestion. We could say, that if you upload a model to a server the default simulation should be a "Minimal Execution Time Scenario". Beside that the modeller should also provide other simulations that would allow to reproduce results published e.g. in a paper.

[llavall]

I suggest that we do it the opposite way - the default simulation is the original one and the "Minimal Execution Time Scenario" needs to be provided and named in a specific way. Following this we can check if the user has defined such a minimal scenario and use it for testing purposes only.

[llavall]

I would also like to add, that we run into performance/security issues if we allow the user to choose parameter in the webportal simulator node, which will lead to a fully blocked server. At least we should think about the different scenarios. One could be Tasjas full_NoV model which runs 7 days on all CPU if the original parameter are selected.

[llavall]

We need a way to select what simulation should be run by name. The idea is to filter for the 'Quick_run' simulation settings and select these to run in testing workflows.

[mfilter]

My suggesion: 1.: we add into our "rule list" a recommendation, that the user should include a "QuickRun_Simulation" into his model. Whenever this model is selected on the server the user will see it and can then decide which one to execute. 2.: We add another "Execution time QuickRun Scenario" to the model repository with execution time estimates for these scenarios, if available 3.: I have now real idea how to prevent users from execution of long running scenarios. For sure this could become an issue

[schuelet]

@miguelalba We have a "writing R code compatible with FSK Lab" section on the wiki, but do we actually have a "rule list" or best-practice list for the users/modellers?