MAVRIC failed to create importance map and biased source

[Zeyu Chen]

Dear all,

         I was trying to perform a shielding calculation for a simplified reactor geometry.

         Previously I have tested the problem using a small geometry (cross-section) of this reactor design. Then moved to the actual 3D full core simulation. 

         However, the simulation returned errors saying: 

         ***Error : BOFWriter attempting to write object for 'ft04f001' failed at offset -1.

         ***Error : AmpxMGBoff failed to write library nuclide at index 1 

         error:  Sequence: mavric failed to run with assertion: Writing Micro AmpxLib failed.

         Can anyone suggest any reason for this? I have attached input/output/msg files.  

 

Thank you,

Zeyu Chen

      

 

[Cihangir]

Dear Zeyu,

I think what you are observing is an artifact of memory management and resource availability and not related to actual MG or BOFF writer. I think your input requires around 500 GB for DENOVO calculations. It is common to use a light-weight (e.g. v7-27n19g library is designed for this) library for the DENOVO adjoint/forward calculations, the MG library after the sequence name (v7-200n47g in your case). I recommend going easy on the DENOVO calculations by reducing the number of voxels in your DENOVO grid (gridGeometry 1) and/or using a coarse MG library for the DEONOVO and switching to a finer MG library for the Monaco using "library".

For example,

=mavric

Monaco-input-study-example

v7-27n19g

.

.

.

read parameters

    library="v7.1-200n47g"

    .

    .

    .

end parameters

I should warn you that it will still require around 20 GB memory with the mesh you have for DENOVO.

I hope this helps!

Sincerely,

Cihangir

 

[Zeyu Chen]

Hello Cihangir,

         Thank you for your reply, that was very helpful. I will try to run a simulation with reduced MG library.

         Is there a way to estimate the memory requirement for input? 

         Previously I was encountering the error "Unable to allocate required memory: bad allocation" at MAVRIC step:4 on a workstation with 16GB ram, so I tried to use "gridGeometry 2" (which contains about 2.8 time less voxels), and the simulation went well until it gave error saying "some source particles leaved importance map".

         So I changed back to "gridGeometry 1" and ran the simulation on our 128GB RAM HPC node and thought it would have sufficient memory.

         How does my input file takes 500GB ram? does that mean part 6 of MAVRIC calculation typically requires much more memory than part 4?

Thank you,

Zeyu Chen

[Cihangir]

What I meant was the execution of Denovo for your input will probably require around 500 GB, not the Monaco MC calculations (requires less).

Denovo (xkba) Memory Requirements:

• For problem parameters 
• Voxel mesh: Nx by Ny by Nz 
• Total groups: G
• KrylovSpaceSize: V
• Legendre order: L
• Quadrature order: S
• Unknowns per cell: N (determined by equation set)
• For a basic xkba calculation
• Memory is proportional to Nx * Ny * Nz * (L+1)^2 * (G + V) * N
•  Extra memory required with first collision 
• (Nx + 1) * (Ny + 1) * (Nz + 1) * G
• Calculation speed is proportional to Nx * Ny * Nz  * S*(S+2)

Cheers,

Cihangir

[Zeyu Chen]

Hello Cihangir,

 

      Actually I thought I was indeed talking about Memory required by Denovo (if I'm correct).

      On the 16GB work station, the error says:

           !*!*! Unable to allocate required memory: bad allocation

           module xkba is finished. completion code 0

      at step 4, and the simulation was terminated at the start of step 6 because lack of a correct "adjoint Denovo flux file"

      However, on the 128 Cluster/node, the simulation ran into error (attached originally with this topic) after about a day (single-thread).

      I guess I'm just curious about what was the difference between these two different errors?  

Thank you,

Zeyu Chen

     

[Cihangir]

As you are already aware "bad allocation" error is about not having enough memory to run a model. The other error, BOFFWriter error, is related to wring an MG library to a file and basically an assertion to be seen/caught by the code developers and not by the users. However, lack of memory/permission/disk space can induce these unrelated errors too.

[Zeyu Chen]

Thank you!

It's very helpful to know that the BOFFWriter error is related to MG library specifically.

Best,

Zeyu Chen