Eigenvalues extraction - Lanczos method

[Sbrabra]

Hi all.

I have a problem during a large eigenvalue extraction (natural frequencies) using the Lanczos method.
The model has about 700.000 unrestrained DOF, and I requested all the eigenvalues under 50 Hz (which means about 200 eigenvectors to be found).
The minimum memory required, as written in the .dat file, is 1.95 GB of RAM (I have 2 GB installed). I monitored the analysis reading the .msg file: the problem requires 5 Lanczos runs and 5 factorizations. Abaqus is able to find all the eigenvectors requested.

The problem is that after Abaqus terminated the extraction, a memory issue is displayed, and Abaqus is not able to write the eigenvectors in the .odb file.
Here is the text of the last rows of the .msg file:

EIGENVALUE EXTRACTION USING THE LANCZOS EIGENSOLVER IS COMPLETE
NUMBER OF LANCZOS RUNS : 5
NUMBER OF LANCZOS STEPS : 99
NUMBER OF FACTORIZATIONS : 5
NUMBER OF BACK SUBSTITUTIONS : 100

***ERROR: STANDARD_MEMORY IS CURRENTLY SET TO 1.95 GBYTES.

***ERROR: ABAQUS HAS BEEN UNABLE TO ALLOCATE THIS MUCH MEMORY. IT IS LIKELY
THAT THE MACHINE BEING USED FOR THIS PROBLEM DOES NOT HAVE
SUFFICIENT MEMORY. POSSIBLE CAUSES OF THIS PROBLEM ARE OTHER
PROCESSES COMPETING FOR MEMORY, INSUFFICIENT MEMORY ON THE MACHINE,
OR A LIMIT ON THE AMOUNT OF MEMORY A PROCESS CAN ALLOCATE. ANOTHER
POSSIBLE CAUSE IS THAT A CONTIGUOUS MEMORY BLOCK OF SIZE 57.50
MBYTES IS NOT AVAILABLE AT THIS STAGE OF THE ANALYSIS.

I don't understand why Abaqus shows this problem AFTER all the eigenvectors were successfully extracted! This memory issue usually appears at the beginning of an analysis, not at the end...
Any suggestion? Thanx to all.

[I work with Abaqus 6.5-1, on a 32 bit Windows XP machine]

 

[xerf]

A suggestion:

" ANOTHER POSSIBLE CAUSE IS THAT A CONTIGUOUS MEMORY BLOCK OF SIZE 57.50 MBYTES IS NOT AVAILABLE AT THIS STAGE OF THE ANALYSIS."


Before running the analysis you should check in Task Manager the amount of available physical memory. Out of the 2Gb some is used by the processes of the operating system. Other programs might run in background as well. In the most fortunate case you probably have ~1.87Gb available immediately after reboot (assuming XP/SP2 starts only the most important services).

Also, ABAQUS/CAE uses plenty of memory for graphical /modeling purposes which is not related to the memory requested to complete the job.

 

[Sbrabra]

Tnx for your answer.

Some considerations on the following analysis (Run A is the one I described in the previous post):

- Run A:

*Lanczos method settings: Requested all eigenvalues in frequency range [0, 50] Hz
*Requested memory (from .dat file): 1.95 Gb
*225 eigenvalues correctly found, but a memory issue made impossible to write the solution in .odb file.
*Max memory really allocated during analysis (from Task manager): less than 1.5 Gb

- Run B:
*Lanczos method settings: Requested 190 eigenvectors in frequency range [0, 47] Hz
*Requested memory (from .dat file): 1.43 Gb
*190 eigenvalues correctly found, but a memory issue made impossible to write the solution in .odb file.
*Max memory really allocated during analysis (from Task manager): less than 1.5 Gb


It's interesting that Abaqus has such different memory requests for two analysis so similar; it seems that memory requests depends on the way you require eigenvectors, not just on the real final number of them.
I monitored the Run B during its course, and I found that the max value of allocated memory occurs DURING eigenvalues extraction, not at the end. So I suppose that the reason why Run B failed at end is due to error "A CONTIGUOUS MEMORY BLOCK OF SIZE 57.50 MBYTES IS NOT AVAILABLE AT THIS STAGE OF THE ANALYSIS.", that I reported in the previous post.

But now I don't understand why the second analysis falied as well. The memory request (1.43 Gb) was compatible with the available memory on my system. It seems strange that the "CONTIGUOUS MEMORY BLOCK OF SIZE 57.50 MBYTES" is the problem again...

I tried another run asking for just 20 eigenvectors: the memory request was 1.41 GB, almost identical to the request for 190 eigenvectors. But in this case the analysis ended successfully.