[overture] Re: 3d cgins memory usage

  • From: Bill Henshaw <henshaw@xxxxxxxx>
  • To: overture@xxxxxxxxxxxxx
  • Date: Thu, 26 Feb 2009 17:44:50 -0800

Joel,
  I suggest using PETSc, but as I noted below there are a
gazillion PETSc options for choosing different solvers,
some of which may not use very much memory at all.

...Bill

Joel Guerrero wrote:

Hi Bill,

According to you, which iterative solver is best to use, the one that comes 
with overture or petsc?.  I will try to run my problem in a clean overture 
instalation, to check if I still have this memory problem.

Regards,

Joel



----- Original Message ----
From: Bill Henshaw <henshaw@xxxxxxxx>
To: overture@xxxxxxxxxxxxx
Sent: Thursday, February 26, 2009 6:06:07 AM
Subject: [overture] Re: 3d cgins memory usage

Hi,

  Just a note that usually all the memory is used up
by the pressure solver and implicit solver, i.e. PETSc
if that is what you are using. You can reduce the PETSc memory
use by decreasing the number of ILU levels etc. and/or
changing the iteration from bi-CG-stab (default) at the
cost of taking longer to solve.

  Also note that if you run with gdb then you can
see who is allocating memory when the heap error
occurs.

...Bill

#DOMINIC DENVER JOHN CHANDAR# wrote:

Hi Joel,

 Generally I have a total of 5 grids.
1 - An outer cartesian (which forms something like a far-field)
2 - An inner cartesian (all sides interpolation )
3 - Wing grid
4 - Wing cap (South)
5 - Wing cap (North)
For Grid (1),(2) I use explicit. Yes I use artificial diffusion. Typical values 1.0.. For nu O(10^-5). I 
observe a change in Cd when I change this
Parameter. The grid sizes that I usually try is around 0.9M-1.3M points. Right now, 
I have a grid with 3M points and I will be
Running it shortly. Will let you know if it crashes... Regards, 
Dominic



-----Original Message-----
From: overture-bounce@xxxxxxxxxxxxx
[mailto:overture-bounce@xxxxxxxxxxxxx] On Behalf Of Joel Guerrero
Sent: Wednesday, February 25, 2009 11:35 PM
To: overture@xxxxxxxxxxxxx
Subject: [overture] Re: 3d cgins error


Hi Dominic,

Regarding this memory issue, I wonder how are you running your
simulations, are you fully explicit? are you using artificial
difussion?.  Im trying to fix my memory problem, but the solver always
crash above 1.3M point.  I checked and the memory is available.  What is
the max grid size that you have been able to sovle?

Cheers,
Joel



----- Original Message ----
From: #DOMINIC DENVER JOHN CHANDAR# <DOMI0002@xxxxxxxxxx>
To: overture@xxxxxxxxxxxxx
Sent: Monday, February 2, 2009 3:31:17 PM
Subject: [overture] Re: 3d cgins error

Do a 'top' in your command prompt to see if the memory used is
increasing with time step. Once it happened that I added a routine that
resulted in a similar situation.

Dominic

-----Original Message-----
From: overture-bounce@xxxxxxxxxxxxx
[mailto:overture-bounce@xxxxxxxxxxxxx] On Behalf Of Joel Guerrero
Sent: Monday, February 02, 2009 10:27 PM
To: overture@xxxxxxxxxxxxx
Subject: [overture] Re: 3d cgins error

Hi,  It seems estrange, I am running on a machine with 8 gigs of memory
so in theory I shouldn't have memory problems.

jg



----- Original Message ----
From: #DOMINIC DENVER JOHN CHANDAR# <DOMI0002@xxxxxxxxxx>
To: overture@xxxxxxxxxxxxx
Sent: Monday, February 2, 2009 3:24:22 PM
Subject: [overture] Re: 3d cgins error

Hi Joel,

I too get this error when there is not enough RAM available. This
usually happens when you run a big test case like yours.
For 1.2M points, it generally uses ~ 4 GB of memory for me. Also, I have observed if you use all boundaries except the inflow as 
"outflow BC", the solution is unstable sometimes. I don't recall when
this happens ( might be when the outer boundaries are quite close to the
body ).

Regards,
Dominic


-----Original Message-----
From: overture-bounce@xxxxxxxxxxxxx
[mailto:overture-bounce@xxxxxxxxxxxxx] On Behalf Of Joel Guerrero
Sent: Monday, February 02, 2009 10:15 PM
To: overture@xxxxxxxxxxxxx
Subject: [overture] 3d cgins error
Hi, I'm trying to run a moving 3d problem with cgins. Usually using a grid 
with less than a million points works fine, but when I use a grid with a
little bit more than a million points (1.2m) I'm getting this error

HEAP ERROR: Array_Data_Pointer == NULL in double* MDI_double_Allocate
(Size = 77616929)! I am using one inflow bc and the rest of the bc are outflow, on the 
moving body I set noslip bc.  Any idea what could be the problem?


Regards,

Joel

Other related posts:

  1. Home
  2. overture
  3. Archive
  4. 02-2009