Preprocessor and multicore : Is is possible?

[Fabi0]

Hi
I'm running abaqus 6.7 on a 4 quadcore HP server with 20Gb of ram (16 core total), the os is Linux CentOS 64-bit.

I need to analyze a rolling bearing with about 25 millions of element (hex and tet): does anybody knows if is possible to use the multiprocessor when writing inputfile? Abaqus uses 16 core only in the secondo part of the job, but pre.exe use a single core only, even if i've checked "multi processor/threat mode" in the jub menu.

With only a single core working, my pre.exe is too slow, it's running till yesterday 6.00am and it don't seem to be finish....

Thx

FABIO

 

[Fabi0]

Just another question...is there, in the world, a preprocessor software supporting multithread?

Thx again

 

[rstupplebeen]

I don't believe PRE can be multi-threaded. I am on 6.9-EF1 and have not seen PRE use more than 1 thread.

Not sure about other pre-processors. I hope this helps.

Rob Stupplebeen

 

[loki3000]

ask simulia about it.

 

[Fabi0]

Unfortunately, Simulia support told that preprocessor only uses one core.

Any idea about multicore preprocessor?

Any of you have ever made a 20million element non linear analysis? ( i was wondering to estimate time needed.. )

THX all

FB

 

[johnhors]

Fabio - an FE solver is able to make use of multiple processors in the solution by taking advantage of matrix partitioning and assigning different partitions to different processors, that is what standard.exe does.

pre.exe as simulia states uses a single processor, but pre.exe is unique to Abaqus, you cannot enter the solution phase with standard.exe without first passing through pre.exe.

If you used nastran (marc) or ansys, it is unlikely whether they would make any greater use of multiple processors than Abaqus, and as you indicate you are running non-linear analysis, each system has it own peculiarities so how would you migrate from abaqus to another system?

Can the size of the model not be reduced? Are 20 million elements really necessary?

http://www.Roshaz.com

 

[Fabi0]

Thx Johnhors, your reply was were helpfull to understand core use!

As a matter of fact, i was trying to reduce my model, but i've a lot of problem because it's a bearing with 104 roller, and Hertz contact need a really small element (about a/6, where a is the width of the contact surface between the inner/outer ring and the roller).

Seeding is also really difficult because the symmetry of the roller.

What i want to do is a wear analisys, so i can't use shell or others...

FB

 

[johnhors]

Can you not use symmetry to reduce the model size?

http://www.Roshaz.com

 

[rstupplebeen]

Sub modeling is also an option.

Rob Stupplebeen

 

[Fabi0]

@johnhors: i've used symmetry, erasing 3 row of roller (bearing was a four raw rolling bearing without cage). The inner and outer ring was reduce to a couple of cylinder (3 mm thickness). I can't do anymore, if a want a realistic simulation of wear, usefull to understand when the bearing is too much damaged.

@rstupplebeen: what do u mean with submodeling?

What really get me angry is that i've a good server, but is useless if pre.exe can't use its 16 core

 

[amubashar]

Hi

Is it possible to post a figure of your geometry? It would help in understanding the use of large number of elements and may be generate any suggestion to reduce the number of elements.

 

[rstupplebeen]

In the "abaqus analysis user's manual" search submodel. 10.2.1 is probably a good starting point. I hope this helps.

Rob Stupplebeen

 

[Fabi0]

Thx for reply, and sorry for delay in answering, but i've been in Turkey for work without any kind of internet access.

Submodeling seems to be very helpfull, but i'm still exploring it, so i can't say more for now

Now i'm trying to get corvengence with only one roll and a small portion of inner/outer ring. (image in attachment)

This is the tipical .msg that i've got from abaqus when i reduced the mesh.....hope someone could tell me more, because the contact interaction I've imposed are with "Adjust to remove overclosure":

"""""""""

CONTACT PAIR (ASSEMBLY__PICKEDSURF21,ASSEMBLY__PICKEDSURF20) NODE
BUR-3-1.36510 IS OVERCLOSED BY 0.167256 WHICH IS TOO SEVERE.
CONTACT PAIR (ASSEMBLY__PICKEDSURF21,ASSEMBLY__PICKEDSURF20) NODE
BUR-3-1.36511 IS OVERCLOSED BY 0.167609 WHICH IS TOO SEVERE.
CONTACT PAIR (ASSEMBLY__PICKEDSURF21,ASSEMBLY__PICKEDSURF20) NODE
BUR-3-1.36512 IS OVERCLOSED BY 0.169069 WHICH IS TOO SEVERE.


***NOTE: SEVERE CONTACT OVERCLOSURES EXIST. CONVERGENCE IS JUDGED UNLIKELY.


***ERROR: TOO MANY ATTEMPTS MADE FOR THIS INCREMENT



ANALYSIS SUMMARY:
TOTAL OF 1 INCREMENTS
5 CUTBACKS IN AUTOMATIC INCREMENTATION
5 ITERATIONS INCLUDING CONTACT ITERATIONS IF PRESENT
5 PASSES THROUGH THE EQUATION SOLVER OF WHICH
5 INVOLVE MATRIX DECOMPOSITION, INCLUDING
0 DECOMPOSITION(S) OF THE MASS MATRIX
1 REORDERING OF EQUATIONS TO MINIMIZE WAVEFRONT
0 ADDITIONAL RESIDUAL EVALUATIONS FOR LINE SEARCHES
0 ADDITIONAL OPERATOR EVALUATIONS FOR LINE SEARCHES
2 WARNING MESSAGES DURING USER INPUT PROCESSING
0 WARNING MESSAGES DURING ANALYSIS
0 ANALYSIS WARNINGS ARE NUMERICAL PROBLEM MESSAGES
0 ANALYSIS WARNINGS ARE NEGATIVE EIGENVALUE MESSAGES
1 ERROR MESSAGES



JOB TIME SUMMARY
USER TIME (SEC) = 10772.
SYSTEM TIME (SEC) = 2719.3
TOTAL CPU TIME (SEC) = 13491.
WALLCLOCK TIME (SEC) = 1182
""""""""