I am trying to generate big 3d meshes (of the order of 500 thousand to 1 million cellls) using GAMBIT 1.2 on NT. My PC has 1/2 a gig of RAM and about 14 gig of disk space and it is a Pentium 450.

I can generate the mesh okay but when it comes to exporting it to FLUENT 5 format or GAMBIT file format, I keep getting problems. The problems I am getting is not repeatable and is very unpredictable. They are

1) Sometimes when I try to export the mesh in FLUENT 5 format, sometimes GAMBIT just sits there and hangs. (I left GAMBIT running overninght).

2) Sometimes GAMBIT give me a mesh but when I try to read the mesh into FLUENT or tgrid, it doesn't work and it gives me an error message saying faces not connected to node etc. Note that I have checked the mesh inside GAMBIT itself and the mesh is okay.

3) Sometimes while I export the mesh, it says that there is memory access violation but it still gives me a mesh which doesn't work.

4) Sometimes it works !

Does anyone have the same sort of problems ? I have a suspicion that it is a memory problem but I cannot be sure. Can any respond to this ?

Thanks.

I used GAMBIT on an INTELLI NT Station from IBM with 768 MB RAM. For generating a mesh of about 735.000 cells I obsereved paging (or swapping). The export somtimes failed like in your case and sometimes not. What I noticed is that saving the dbs and starting a new gambit before exporting mostly helped. It seemed to be a memory handling problem of NT.

Yes, It seems to be a memory problem. You may get it confirmed by sending the model to your Fluent support engineer. Also, exporting the mesh as Gambit neutral file will require less memory. You can later convert it into Fluent mesh format, by using Fluent.Inc/bin/tfilter fe2ram from outside the Gambit, or can directly import the Gambit neutral file in Fluent.

Any plans of correcting this error in the near future ? It looks like many people are having this problem with GAMBIT (see previous post by Volker Pawlik).

I am actually a big supporter/fan of GAMBIT and I am trying to convince the peoeple in my organisation to convert from ICEM to GAMBIT. But it is very difficult to do that if they know that the above bug is present or come across the problem in the mesh that they are trying to generate !!!!

Are you joking?

Converting from Icem to Gambit for heavy problems seems to be the worst possible idea.

This does not seem to be a Gambit bug, but a system limitation where user's machine does not has enough memory to export such a big mesh. However, if Volker feels it could be a bug, he should send the dbs file to his Fluent support engineer. We need the file to reproduce it and if confirmed, will fix it as soon as we can.

Moreover, we also have plans to make Fluent 5 mesh export more memory efficient in general.

Okay, let's try it out on a test problem. I've created a journal file below that generate a mesh with a million elements for a 10x10x10 brick.

---start of journal file----

/ Journal File for GAMBIT 1.2.0 / File opened for write Tue Jan 25 10:23:32 2000. default load "C:\users\default\GAMBIT.ini" volume create width 10 brick volume mesh "volume.1" map size 0.1

----end of journal file------

and then I export the mesh as a gambit neutral file. When I try to import the mesh into FLUENT 5.2.3 I get the following error message.

-----start of message--------

Welcome to Fluent 5.2.3

Copyright 1999 Fluent Inc.

All Rights Reserved

Loading "c:\fluent.inc\fluent5.2\lib\fl_s1185.dmp" Done.

> Reading "| tfilter fe2ram -cl -tGAMBIT -cl -d3 test.neu"...

GAMBIT file to RAMPANT file

1030301 nodes. 2970000 quadrilateral interior faces, zone 3.

60000 quadrilateral wall faces, zone 4. Read 1030301 nodes, 1000000 elements, 1 groups. 1000000 hexahedral cells, zone 1. 1000000 hex elements

Done.

Building...

grid, face[1863461]->node3[8812453] out of range (0<node<1030301). Error: Build_Grid: grid error.

Clearing partially read grid.

Done.

------end of message------------

Again, I remind you that I am running GAMBIT 1.2 and FLUENT 5.2.3 on an NT workstation with about 20 gig of diskspace and about 0.5 gig of memory. We have tried the exact grid using ICEM and importing it into FLUENT on the same computer with no problems so I very much doubt the problem is with FLUENT or NT or the "limitation where the user's machine does not have enough memory....". If Icem can export a mesh on the same machine, it will not be unreasonable to expect the same of GAMBIT.

Again I remind you that I am supporting gambit. I am just trying to convince people in my organisation to use it but it is very difficult if we keep encountering this problem. For the problems that we are trying to solve, we need big meshes and we can get them with ICEM on the same machine....but we prefer to use GAMBIT.....if we can !!

I've also had the same problem exporting most of the Fluent mesh's (V4.5 or 5) from Gambit. But by making it a generic neutral file they often seem to load fine except for the larger one's. I'm running twin 550Mhz Zenon's w/ 1/2 gig of ram and when problems generally appear, they seem to occur when writing to the hd. I've also thought it may be a hardware problem in the ram and have been looking for a way to test the modules.

I'm glad to hear that you are having the same sort of problem. It is comforting to know that I am not imagining things. Let's hope that we have enough momentum now to obtain a response from FLUENT.

To be fair, I ran a test case once, both on an NT workstation and a SGI machine with the same amount of memory. I had problems with the NT machine but I managed to get a grid without any problem on the SGI workstation. So it must be something to do with NT.

If I run a bigger problem that uses up all the memory on the SGI workstation, at least I get an error message saying that it cannot allocate memory to face.... When I run out of memory on the NT machine, sometimes it just hangs, sometimes it gives me a grid that doesn't work.....with different error messages each time !

When generating large meshes in Gambit I always try to do small parts separately and then merge them together with "tfilter tmerge3d". Gambit runs so much better on small problems. You can mesh one part, delete the volume mesh and only keep the mesh neigbouring to other parts and then continue with those.

I run Gambit on a high-end HP with 2 gigs of RAM. It starts to get slow when you hit a million cells and when you are up to 2 or 3 millions you run into real problems with intermittent hang-ups etc.

Hello, no I do not think, that it is a bug of GAMBIT but an NT specific problem. An NT admin told me that NT servers e.g. do have to be rebooted once or less a month just due to the fact that previously allocated memory is not released. That is what I supposed to be the problem.

Andrew,

You are hitting memory limits on your NT machine. As Shyam Kishor suggested, change your solver to "Generic" and export your mesh as a Gambit Neutral File. You can then import this file into Fluent 5.x.

Shyam also correctly pointed out that we will be implementing a more memory-efficient Fluent 5 export later this year.

Scott,

Thank you for the information.

Sincerely,

A. Ooi