Learning Forum

[shezanghamat]

I read a .msh file in fluent. I converted the tetrahedron mesh elements to polyhedra and set up case settings. Before initialization I tried to save the case file but I got the following even though volume had enough space.

Error at Node 0: Buffered write, in function allocate_mem_buffers: unable to alloc memory (3251617 bytes) for mem_bufferp->mem_buckets[i].

Error at Node 2: Buffered write, in function allocate_mem_buffers: unable to alloc memory (3251617 bytes) for mem_bufferp->mem_buckets[i].

Error at Node 3: Buffered write, in function allocate_mem_buffers: unable to alloc memory (3251617 bytes) for mem_bufferp->mem_buckets[i].

.........................................

Error at Node 21: Buffered write, in function allocate_mem_buffers: unable to alloc memory (3251617 bytes) for mem_bufferp->mem_buckets[i].

If I save a case file right after converting my tetraderon mesh to polyhedra, I am able to save it correctly. But as soon as I set the case settings and then try to save it, it shows the above error. I even tried on different system but I got the same error so I believe It was not the system issue. I am unable to understand what exactly the issue is. I sincerely ask for your assistance. Thank you for your time.

[Rob]

Did you convert to poly or directly create the poly cells? Are you in Workbench? 

[shezanghamat]

Mesh was created in tetrahedron elements. I converted to polydehra in fluent. No I am not in workbench, I'm directly opening fluent by setting a working directory there and importanted the tetra mesh file there itself.

[Rob]

We'd generally generate the poly mesh directly now - it's usually better quality and more controllable.  The conversion function should still work, how many tet cells are there, how much RAM and how many parallel cores are you using?

[shezanghamat]

Understood your comment, I will try to create it directly in polyhedra. The issue just got resolved. It was probably RAM issue. Just for your reference, mesh is of around 56.2M tetra cells and 9.84M poly cells after conversion. I was using 98GB RAM initially, now I switched to 128GB RAM. I am facing a new issue now, once I start the solution, it is taking too long to start with the calculation. Its been more than half an hour and not even a single iteration is solved. I am unable to identify whether the fluent is freezed or its solving.

[Rob]

A native poly mesh of 10M cells ought to fit on 98GB RAM but the tet mesh would be very tight, and I'm not sure about the tet-poly conversion. How many cores are you running on? Fluent tends not to stall but if you're out of RAM (128MB is marginal for 56M cells) you may be paging to disc and that can take ages. 

[shezanghamat]

I am using 22 croes. I checked after the starting the calculations 109GB RAM was in use.

[Rob]

How many physical cores, and how many memory channels? 

That'll explain why the smaller machine failed, and you're very close on the larger one. How does the load on the cpu's look? 

[shezanghamat]

Sorry for the delayed reply. CPU utilitzation was 71%. I have total physical cores 24 and logical processores 32. I am not sure about number of memory channels. Can you specify more on that?

[Rob]

Memory channels is how many routes data has to pass between the nodes: most high core count chips don't have enough to pass the amount of data that Fluent moves around in a parallel case. Can you check if each core is around 70% load or if you have some at 100% and some idling? I suspect you've just run out of compute rather than having a problem in the model. 

[shezanghamat]

Understood. I will have to run it again to check that. Will get back soon. Thank you.