Change in burn behavior due to mesh partitions

[sebastianmezaurotadze]

Id like to run simulations on very large domains (millions of cells) and use many partitions (~64) in order to speed up the compute time. I however notice that using a different number of subdomains (i.e. 64 vs 32 vs 16) can change burn behavior significantly.

Any Insights into the following would be helpful:

• How information is transferred across mesh boundaries
• Why does using more subdomains change the simulation behavior
• How to mitigate this error from many domains
• How to accelerate computation of large domains

Thanks

[rmcdermo]

Have a look at this chapter in the user guide: 21.1 Accuracy of the Pressure Solver

[mcgratta]

Can you post a plot of the two different heat release rate histories and the input files that generate them?

[sebastianmezaurotadze]

Found that lag in radiative heat transfer was likely the culprit.

Here are some sample tests on smaller domains with HRR graphs comparing a single domain / split domain / split domain with higher radiation iterations to transfer radiative heat

FDS Code:

Single_1x_Domain_1x_Radiation_Iteration.txt
Split_4x_Domain_1x_Radiation_Iteration.txt
Split_4x_Domain_15x_Radiation_Iteration.txt

[sebastianmezaurotadze]

Also, running a simple test with a flux gauge and a radiative particle separated by multiple mesh divisions on the default FDS settings shows up to a 3 second delay even over short distance (2.8 m or 8 cells)

Any recommendations on how to optimize RADIATION_ITERATIONS as to maintain simulation accuracy without excessive runtime?

[mcgratta]

RADIATION_ITERATIONS is 1 be default. However, in cases of multiple meshes, you might want to increase it to 2 or 3 if it is very important that the radiation field be updated over the entire domain. Radiation intensity is exchanged at mesh boundaries once per solve. That means that a remote mesh does not "see" a heat source until FDS runs through X number of solves. Try setting it to 2 or 3 and see how that affects your problem.

[mcgratta]

Any progress on this?

[sebastianmezaurotadze]

Increasing radiation iterations helped, it didn't slow down my simulation significantly. This thread is already resolved.