search

EarthCubeInGeo / resen-core

The core docker image used by resen.
GNU General Public License v3.0
2 stars 3 forks source link

[DISC]: Using Models in Resen #58

Open ljlamarche opened 4 years ago

ljlamarche commented 4 years ago

This thread aims to foster discussion on the possibilities and challenges of adding geospace models to Resen. The InGeO team would like to continue the conversation that was started at CEDAR 2020 with numerical modelers about how Resen and containerization in general can be used to assist with computational reproducibility in the modeling community. In addition, we would like to know if packaging models in this way can help make them more accessible to a wider range of researcher and students. We welcome thoughts and comments, and are happy to answer any questions about Resen!

mattzett commented 4 years ago

I think a useful application of Resen that we haven't brought up yet would be a situation where you are using a numerical model to illustrate something in an educational setting. In this case having an easy way to deploy a working example to a bunch of students and allowing them to tweak the inputs and look at the results interactively could be quite valuable. Normally for GEMINI this would require building a bunch of libraries and waiting maybe up to ten minutes for everything to compile. That's just too much to normally ask of a class full of physics students. By the same token a person who is just starting to learn a model could begin working with it this way instead of spending a few hours trying to build everything from scratch. Our build system tries to deal with this but the fact of the matter is it still takes quite a bit of expertise to use.

mattzett commented 4 years ago

Regarding use for reproducibility, resource management is a major consideration. Pardon my relative ignorance, but does running something with Resen incur substantial overhead? Can it be made to interact with an HPC queuing system? A lot of our stuff gets done on multi-node systems these days so if someone wanted to quickly reproduce a result they'd need to run on similar hardware (or wait a long time)...

ljlamarche commented 4 years ago

Using Resen in an educational setting to give students easy access numerical models (even just "toy" models to illustrate concepts) is potentially very useful. One idea is for instructors to build custom cores that include not just the installed model but several example runs and relevant tools, maybe to be used in conjunction with a class. We're hoping to enhance the Resen tool so that it can handle custom cores, but even requiring students to get python and docker installed on their computers is again probably a bigger barrier than most instructors want to deal with in a classroom setting. The online system was designed to really let people get going with software as quickly as possible, which I suspect what you want in a educational setting. We've talked about setting up the online system to also have a variety of specialized cores users could choose from (so students would be able to log in and select the "gemini-core" and have that environment immediately available), but I'm not sure our current system can handle multiple accounts running even highly simplified numerical models. I think server resource allocation would have to be handled very carefully for something like that.

asreimer commented 3 years ago

Note: We have had success in installing some models. See PR: https://github.com/EarthCubeInGeo/resen-core/pull/72