Interest in Community Led RAS 2D Experiment?

[Anonymous]

Hey Everyone,

I would like to try to set up and run a RAS 2D run performance experiment. The goal of this would be to take a few 2D RAS models with various configurations and run them on as many machines as possible to try and build a performance profile. Depending on how many people are interested in participating, we can start to answer easy questions like “About how long will this model take to run on my hardware (aiming to hit within 10 minutes of actual time)” to perhaps more interesting questions like “What maximum time step and cell size will still generate a stable model”?

Sky’s the limit really. Just depends on level of interest and participation.

To get started, I have set up a google drive folder for volunteers to drop test models in for everyone to run. Please don’t put anything in that folder you don’t want everyone else to see as everyone will have access to it. Also, please don’t nuke it with large files. I only have a free Google Drive account at the moment. =]

Link to Drive Folder

For the basics and to get solid run time stats we will need to run a few models (1-3 i think) on a lot of machines.

If you can’t submit your model but would like to submit your basic configuration and run stats for the community database then I will make sure you have a place to do so in the Google Form I am working on.

If you’d like to submit your model for community use please adhere to the following guidelines. This will help me facilitate and make any normalization changes for the study.

Submission Guidelines:
+Submit a model with a “Base” run that does in fact run (This will be very important with models with massive grids). Please make sure it can run through completely without problems.
+Clean up your model by removing unneeded plans, layers and etc… It needs to be easy to understand by everyone.
+Make your .dss, flow, geometry, plan and other RAS file names explanatory and or easy to understand.
+Please include a projection file for your RAS model
+Please DO NOT include full geoTiff or similar DEM files for ground geometry and land use. Just the HDF files RAS creates for computations should suffice.
+Please include a text file describing your model and anything unique about it that might be interesting, hard to understand or might be significant to computation time. +Also, add the following information to your description text file:
+DEM resolution
+Dx/Dy Value for grid generation
+Average slope in study area

I haven’t quite figured out exactly all that I want to put into the 2D RAS survey form but when I do I will need to be able to understand your model enough to make cosmetic changes to fit the study parameters.

I hope a few of you are interested. For more details on the project see the link here that will take you to a community Google Doc. Everything is open at the moment so hopefully that goes well. If it gets too crazy I will need to lock it down and make it read only. Feel free to post comments, suggestions and or tell me if I am doing something horribly wrong. Thanks!

[cameron]

The FMA Modeling and Mapping Committee has just been doing this, but in the cloud and for a range of software not just HEC-RAS. The basic idea was we setup a cloud that people could remote into and test different types of software for three different datasets along with a few different machine configurations. The results of the findings are going to be presented at the FMA conference in San Diego in September.

We started working on this idea last year and was a big effort by those on the committee and I hope everyone at the conference enjoys panels.

Side note – none of your links worked

[TypicalEngineer]

Thanks for the heads up about the FMA group. I haven’t heard about that or the effort. Please share any links if you have them. It would be great to skip all the boring stuff and go straight to interesting with the FMA results available.

Question: When you say different machine configurations were these VMs also in the cloud? For this to be really useful a lot of different configurations would need to be tested. What exactly was the goal? Was it a typical cloud architecture or was there hyper-convergence or any GPU computing?

Thanks for pointing out the links! Looks like copy and paste doesn’t take well on this forum.

Folder to Drop Models: https://drive.google.com/open?id=19VZqVt_GIqYKDAtZOGx-wReMevCsJqzN

Google Doc: https://drive.google.com/open?id=1Y12pM89ZaR_hfVUXNpp7GMYYcSY8jRmJcYk56_ivSOY

[Author]

Posts