Hi everyone,

we are two students from berlin working on a cfd simulation of a wind turbine. We finished grid generation an some low resolution(1 Mio. Hexas) calculations. Now we try to set up our final simulation with approximately 3 Mio Hexas. We use 3 Computers everyone with 2GB of RAM (Windows XP,CFX 5.7.1,SP2). So there is about 4.8-5 GB Free RAM available. When we start the simulation with the solver, its quits with "Run-time memory configuration error - Not enough free memory is currently available on the system.Could not allocate requested memory - exiting!". This happens when the solver tries to allocate the memory for partioneering. So why is there not enough memory for partioneering a 3 Mio Hexa (Multigrid)with approx. 5 GB of RAM ? Thanks Kramer

I met the same experience about the memory problems. Try to set two parameters-"memory allocation factor"for solver and partition respectively.Default value is 1.0, try 0.9,1.2,1.4 and so on,

Hi Fred,

thanks for the help, i tried to set the memory allocation factor but it still wont work. I have heard from a friend that the problem is that under windows no parallel partioneering is possible. So we have to partioneer on some linux machines and then go back to our windows computers to start the solver. Do you believe that it is impossble to partioneer big grids with windows or do you know something about it?

Thanks Kramer

If you have Linux on your machines, it will be better for it. You will get more use of your memory.

Due to system limitations on how the partitioner allocates ram, the maximum memory allocation on Windows for the partitioning is 1.7 GB (if you are lucky) which translates into about 0.425 Gwords of solver space (divide by 4 because the flow solver allocates 4 byte integer and reals) to be divided between all the different data types.

I suggest you use one of the "cheaper" partitioning methods instead of the default, MeTiS algorithm. Try optimised recursive bisection or standard recursive bisection for example.

Playing with the individual integer and real memory stack multipliers might help. The partitioning step is integer stack limited. The maximum you can use is some fraction of the 425 Mwords above. You might get away with 300 or so.

"Cheap" partitioning methods probably need about 25 words per node for structured grids. So for your case you need 75 Mwords or so. So, I think it should fit.

Neale