search

lautenberger / elmfire

Eulerian Level set Model of FIRE spread
https://elmfire.io
Eclipse Public License 2.0
30 stars 13 forks source link

Simulation stop halfway before reaching TSTOP #32

Closed dwimjpurnomo closed 1 year ago

dwimjpurnomo commented 1 year ago

When I run for 30hr, most of the time it only gives the result until 15hr. The input deck is the following.

https://drive.google.com/file/d/1YtfXC3v-L7JPomIBeyAVtQoGsqkjhsWE/view?usp=sharing

lautenberger commented 1 year ago

Thanks Dwi, I'm looking into it

lautenberger commented 1 year ago

Dwi- I launched this run 2 days ago and it's still chugging along, albeit very slowly. I don't think there's a bug here, I think it's just taking a very long time to run.

I had ELMFIRE dump intermediate timings and about 99.99% of CPU time is spent in the urban spread routines. Let's spend some time to get these run times down when urban spread is enabled, it will pay off in the long run. Have you done any profiling of the urban spread routines with Intel Vtune Profiler or similar?

dwimjpurnomo commented 1 year ago

Yes, Thomas fire was very long even with Hamada, so it will be even longer with the new model. I will try Intel Vtune Profiler.

dwimjpurnomo commented 1 year ago

I am currently using a virtual machine for vtune, because my PC has windows OS. Attached is the message that I cannot have analysis on memory access. Is that OK? vtune

lautenberger commented 1 year ago

Hi Dwi- I just checked and this case ran to completion locally. It took about 3 days, Sep 9 - Sep 12. So I'm going to close this issue and let's move the optimization/profiling discussion over to Issue #16. Based on the above screen shot It seems like Vtune is saying it can't do GPU hotspots which doesn't matter since ELMFIRE doesn't use GPU for anything. Not sure about the memory access failure though.