DEVC "GAUGE HEAT FLUX GAS"

[rupole1185]

Describe the bug
The following fds simulation has a two different behaviour: if the DEVC "GAUGE HEAT FLUX GAS" are present in the file the simulation is about 50% slower than the same simulation without these devices

To Reproduce
Run the analysis with and without the DEVC "GAUGE HEAT FLUX GAS" on FDS 6.9.1 and FDS nighly

Expected behavior
I wonder if such a slow down is normal or it could be reduced. What I notice is the CPU usage decreases with time in the simulation with DEVC, while it does not decreas with time without DEVC

Desktop (please complete the following information):

• Version: FDS 6.9.1 and nightly

TestGAUGE_TestGAUGE_Cleaned.fds.txt

[Er9y714]

Did you compare the timestep and number of iterations?

[rupole1185]

In two example I have made I noticed that they got to 900 iterations (46.38s) in just two different time.
With the gauge it took it about 30 minutes, with without it it took approximately 15 minutes.

Time step number and simulation time seems not to change between the two cases, it is just the computational time to get there which is higher

[mcgratta]

I cannot run the case. There are vents that are not attached to any surface.

ERROR(820): VENT West Test Exhaust_01 has no solid backing or an orientation index (IOR) is needed. (MPI Process: 7, CHID: Test_GaugeDevice_with)

[rupole1185]

Hi Kevin,

yeah sorry, you have to remove the following 6 vents:

Beside that, I have retested it with today's nightly and still the slow down when GAUGE devices are present is measurable. The simulations are about 20% slower with these devices.

[mcgratta]

OK, I'll run the cases. My first guess is that by using these "_GAS" devices, you are creating some overhead having to do with particles. Any "_GAS' device is a particle that records information.

[drjfloyd]

In read at 5792 we have:

IF (TARGET_ONLY) THEN
...
SOLID_PARTICLES = .TRUE.
...
ENDIF

Should we be setting SOLID_PARTICLES=T here? This causes us to evaluate particle interactions with radiation, but a TARGET_ONLY particle isn't meant to exchange energy with the gas, just be able to record data. However, the amount of extra code tied to this doesn't look like enough though to give a 20+ % change.

I deleted all but the OPEN VENTs and ran for 10 s with and without particles on our cluster and the runtimes were 526 (no particles) and 543 (with particles) or 3.3 % slower. Ran for 30 s and results were 1542 and 1556 or 0.9 % slower.

[mcgratta]

I ran the case for 100 s. The CPU usage was 5288 s for the case with the heat flux devices and 5247 s without.

The extra 40 s was spent mainly in the radiation and output routines. This is not too surprising. Memory usage is about the same

[mcgratta]

By the way, you can see the timings of the major subroutines in the file job_name_cpu.csv

[mcgratta]

Could you post the _cpu.csv files for the cases that show a large difference in CPU time.

[rupole1185]

Hi there, sorry but I have had busy days before Christmas. Well, I should re-generate the CPU files ... I'll try to do that after Christmas! Merry Christmas to everyone! *Ruggero Poletto - **Managing director* *VideoCall*: calendly.com/ruggero-poletto | *Mobile*: +39 378 30 33 133

…

________________________________________ CFD FEA SERVICE SRL P.IVA 04545570238 - via Borgo Grande 19, 37044 Cologna Veneta VR *PRIVACY and statement* All information in this e-mail message, as well as any files included, are intended for the exclusive use of the recipient (s) indicated above / indicated, and therefore subject to protection pursuant to law no. 675 and of the Legislative Decree of 30 June 2003 n. 196. Anyone who receives this message by mistake, is requested to notify the writer of the incorrect receipt, and proceed with the immediate cancellation of the same from the system. Your cooperation is appreciated. Check our privacy policy here <https://cloudhpc.cloud/privacy-policy-2/>.

On Tue, 17 Dec 2024 at 22:22, Kevin McGrattan ***@***.***> wrote: Could you post the _cpu.csv files for the cases that show a large difference in CPU time. — Reply to this email directly, view it on GitHub <#13866 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABXMUCJNZAGHB635M7KFX6D2GCI2RAVCNFSM6AAAAABTLICR4SVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNBZGY3DCNZRG4> . You are receiving this because you authored the thread.Message ID: ***@***.***>