 Re: crash on blade center system ( No.1 ) |
- Date: 2005/05/31 10:24
- Name: T.Ozaki
- Hi,
> The openmx 2.73 always kill nodes when memory overflow invoked.
Although I am not familiar with your computational environment,
I am wondering that it would be possible to assign virtual memory
by specifying environmental variables to avoid such a situation.
> Please, let me know how to estimate memory, before calculation,
> even it is crude method.
It is not so easy to estimate the memory requirement before calculations.
A simple way is to increase the system size and parameters step by step
and to know how the memory size increases using a system command such as
"top" as a function of those parameters.
> Also, I want to know this program has any limitation of cpu numbers.
The maximum number of cpus you can use is equivalent to the number
of atoms in your system.
Regards,
T.Ozaki
|
| Re: crash on blade center system ( No.2 ) |
- Date: 2005/06/01 01:19
- Name: Sergey Lisenkov <proffess@yandex.ru>
- Dear Chong Sang Hyuk,
It seems your jobs requires a lot of memory, much bigger you have. I succesfully compile and run OpenMX code on our IBM JS20 PPC970 cluster (OS SuSE Linux and gnu compilers (xlc fails to compile a few files)). But we have 4Gb memory on each node, so we have no any problems with memory allocation... Because as I understand it is impossible to reduce memory requirements by increasing the number of nodes, you might use the code on small jobs.
Best wishes,
Sergey
|
| Re: crash on blade center system ( No.3 ) |
- Date: 2005/06/02 08:57
- Name: Chong, Sang Hyuk
- thanks a lot for yours answers
It helps to me !!
|
|
This thread is locked.Only browsing is available.
crash on blade center system