Re: Memory leak using trilinos

[Zhekai Deng]

Hi Mike,

Thanks for your reply. I just tried the garbage collection and it seems did
not help very much with the leak issue for trilinos. My Pysparse throw me
Segmentation fault, which I don't really understand why.

For the check point approach, just to make sure that I understand it
correctly, are you using dump from Fipy.tools to store the solution
variable, and read them back ? I guess I also need to re-declare my
equation formulations, and boundary conditions each time I read them back.
Also, I wonder how to delete/remove all the previous object so that I could
have a "fresh" start after each check point iteration ?

Thanks,

Zhekai


On Tue, Jun 20, 2017 at 4:51 PM, Michael J. Waters 
wrote:

> Hi Zhekai Deng,
>
> I had a similar problem. In the end, I just created check point files and
> restarted frequently.
>
> I did find that garbage collecting helped slow the leak and,
> interestingly, also sped up code execution for me.
>
> Best,
>
> -Mike
>
> On 6/20/17 10:39 AM, Zhekai Deng wrote:
>
> HI All,
>
> I think I run into some memory leak issues using trilinos as my solver
> both for serial and parallel case. The total memory goes up from initial
> around 300 MB to the my system memory limit (like 29 GB), and program
> stops. I have attached my example code below.
>
> Somethings I am not sure if I implement it correctly:
> 1. in the example code, I basically solve multiple equations at the same
> time. I use list, and append my CellVarible together
>
> 2. Similar, I use list and append my equation together
>
> 3. Create coupled equation by looping through the list element, and solve
> this coupled equation.
>
> 4. During the solution step, I loop through the list element, and do
> updateOld()
>
> The results of the numerical calculation looks correct to me, I just have
> this memory issue that prevent me from going longer time steps because I
> run out of  memory.
>
> I wonder is there any workaround/fix to this problem ?
>
> I am using the build from Anaconda repository, default trillions solver,
> and my OS is Ubuntu 16.04.
>
>
> ___
> fipy mailing listfipy@nist.govhttp://www.ctcms.nist.gov/fipy
>   [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]
>
>
>
> ___
> fipy mailing list
> fipy@nist.gov
> http://www.ctcms.nist.gov/fipy
>   [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]
>
>
___
fipy mailing list
fipy@nist.gov
http://www.ctcms.nist.gov/fipy
  [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]

Re: Memory leak using trilinos

[Michael J. Waters]

Hi Zhekai Deng,

I had a similar problem. In the end, I just created check point files 
and restarted frequently.


I did find that garbage collecting helped slow the leak and, 
interestingly, also sped up code execution for me.


Best,

-Mike


On 6/20/17 10:39 AM, Zhekai Deng wrote:

HI All,

I think I run into some memory leak issues using trilinos as my solver 
both for serial and parallel case. The total memory goes up from 
initial around 300 MB to the my system memory limit (like 29 GB), and 
program stops. I have attached my example code below.


Somethings I am not sure if I implement it correctly:
1. in the example code, I basically solve multiple equations at the 
same time. I use list, and append my CellVarible together


2. Similar, I use list and append my equation together

3. Create coupled equation by looping through the list element, and 
solve this coupled equation.


4. During the solution step, I loop through the list element, and do 
updateOld()


The results of the numerical calculation looks correct to me, I just 
have this memory issue that prevent me from going longer time steps 
because I run out of  memory.


I wonder is there any workaround/fix to this problem ?

I am using the build from Anaconda repository, default trillions 
solver, and my OS is Ubuntu 16.04.



___
fipy mailing list
fipy@nist.gov
http://www.ctcms.nist.gov/fipy
   [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]


___
fipy mailing list
fipy@nist.gov
http://www.ctcms.nist.gov/fipy
  [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]