What effects FDS RAM usage?

[norm07]

Hi All,

I've recently been running models where I've mucked around with sprinklers and its blown out my RAM usage. Mesh size is around 200,000. However the sprinklers are running at 339 l/s, constant particle diameter, particle count of 10,000, velocity of 18.1.

What settings in general (besides mesh) cause a dramatic increase in RAM usage?

Kind regards,
N.

[drjfloyd]

Are you using the current release of FDS? If not try that. We keep tweaking how particle memory is managed (it needs special treatment since particles can move between meshes and hence between different MPI processes).

By particle count of 10,000 do you mean the number of particles per second per sprinker or the total number of particles? If it is the particle per second from the sprinkler, then you might just be accumulating a lot of particles in your domain. You could limit the particle age or maximum number of particles.

[drjfloyd]

The only thing I can think of is fine watermist at operation where drops might evaporate in the first time step. In that case, if evaporation was first then there would be no droplet drag when it came time to do movement. Unless we kept track of the old and new radius. Then we could impose drag as a time average. If we assume a constant evaporation rate then the old and new radius would give a function of radius over the timestep and we could average that radius squared over the timestep for drag.

[norm07]

thanks for looking into this guys. I will leave this in your capable hands for the time being. Just wondering, is there a work around for this at the moment?

[drjfloyd]

One issue was fixed today. If that passes firebot, then tomorrow there will be a test bundle at https://github.com/firemodels/test_bundles that you could try. That might help. If not then the other issue related to Q_DOT_RAD has a likely fix but that might not make it into tomorrow's bundle.

[norm07]

forgive my ignorance of the back end software stuff but are you saying i should wait for an updated FDS release to see these changes come through?

[drjfloyd]

Every night an automated tool called firebot checks out a fresh copy of the fds and smokeview git repositories, compiles fds and smv, runs a suite of ~900 test cases, checks all the results against known values, and builds the documentation. If all steps are successful, it then creates test bundles that are put on the github site in my prior post. You can look at the website tomorrow morning (Washington, DC time) and look to see if there is a new bundle for fds. If there is, you can download and install that bundle and try using it for your case. If that doesn't work then the next set of changes might but those are not comitted yet for firebot to test.

[norm07]

Hi Doc,

I've been having a great time modelling sprinklers following the update. Until I ran into some trouble with insufficient RAM.

Please see attached FDS file. I get the error at around 400 seconds (5th sprinkler activation from my view), do I simply just not have enough RAM or is this a similar bug as previous?

File attached.

Kind regards,
Norman.
222150_DF1_SEN.txt

[drjfloyd]

What is the error message you are getting?

[norm07]

insufficient virtual memory. In addition to the above model, I was also running a model of a similar size at the same time. My system has 32 Gb of memory.

[drjfloyd]

You may not be able to run two cases at once with your current inputs. With 5 sprinklers operating and 10,000 particles per second per sprinkler, you are going to have a lot of particles to track. If you are on a windows machine, depending on how much stuff you have running in the background you could be loosing 8-10 Gb just for the OS. You coud try reducing the particles per second. You could make a case where you just start a sprinkler right away with TIME as the QUANTITY and look at how many particles per second you need to get a decent looking cascade of particles down the boxes.

[norm07]

Thanks for your input doc, ill muck around with it for a bit and see what works ... or spend some money on some RAM.