Incorrect Number of Entries for Vector Received

This forum is for issues related to parallel processing
and OpenSees using the new interpreters OpenSeesSP and OpenSeesMP

Moderator: selimgunay

Post Reply
AHARSHIAN
Posts: 87
Joined: Tue Apr 03, 2012 3:34 am
Location: Germany, Weimar
Contact:

Incorrect Number of Entries for Vector Received

Post by AHARSHIAN » Wed Oct 22, 2014 4:02 am

Hi,

I am running a relatively big model in OpenSees-SP. Sometimes I get the following error. I would like to know the reason, and How can I fix it?

Error: Incorrect number on entries for vector received: 2 expected: 8550

Also, a window pups up which says that your computer is low on memory.

I am running on a 24 cores machine with 16 GB RAM. System: Windows 64-bit

Another question: For a given FEM, how can I estimate the optimum number of cores that I have to work with for reducing the simulation time?

Thanks for Your Help....
Best Regards,
A.H.Arshian

fmk
Site Admin
Posts: 5883
Joined: Fri Jun 11, 2004 2:33 pm
Location: UC Berkeley
Contact:

Re: Incorrect Number of Entries for Vector Received

Post by fmk » Thu Oct 23, 2014 8:27 am

that error if it pops up without that system memory error means there is a bug in the code .. if popping up with the sys low error it could instead be due to running out of memory and program is not catching tit

as for the #cores, speedup is a function of solver time, element state determination (and memory issues if all the processes cannot keep their data in the RAM you have available) .. typically for high rise buildings after 8 proceses the solver starts slowing down (though this can happen after 16 or 32 if a continuum model of a brick say) .. the element state determination scales perfectly, but it's influence is typically minor as most time spent in the solver .. the one that can pop up all of a suddden on a multi-core is the page faulting .. when you add more processes to an SP computation there is more memory overhead, e.g. each processor needs vectors for the solvers X and B,

AHARSHIAN
Posts: 87
Joined: Tue Apr 03, 2012 3:34 am
Location: Germany, Weimar
Contact:

Re: Incorrect Number of Entries for Vector Received

Post by AHARSHIAN » Thu Oct 23, 2014 8:38 am

Hi Frank,

Thanks for your reply,

The fact is that the code with its all elements, sections, load patterns, and almost everything related to the analysis part is working perfect for a 2D frame. But then when I apply it to a 3D model, such a problems occurs.

I am running two analysis in series. First a nonlinear static and then a nonlinear dynamic. For some reason, I cannot use Mumps for the static case and I use SuperLU. For the dynamic I use Mumps, but I had some problems and I fixed it with using ICNTL14 option (-ICNTL14 100).

Something is really strange. As the model has to be ran for 1000 times, the problem I talked about happens after a couple of analyses. I mean that fore example the code works well for 10 analyses, and at the 11th analysis where just material properties are changed suddenly this problem occurs.

Thanks for your help.
Best Regards,
A.H.Arshian

fmk
Site Admin
Posts: 5883
Joined: Fri Jun 11, 2004 2:33 pm
Location: UC Berkeley
Contact:

Re: Incorrect Number of Entries for Vector Received

Post by fmk » Thu Oct 23, 2014 1:49 pm

are you doing the multiple analysis from the 1 script .. or starting them seperatly .. if from 1, it could be you have discovered a memory leak

AHARSHIAN
Posts: 87
Joined: Tue Apr 03, 2012 3:34 am
Location: Germany, Weimar
Contact:

Re: Incorrect Number of Entries for Vector Received

Post by AHARSHIAN » Fri Oct 24, 2014 12:42 am

Hi Frank,

Yes, I am running all in a single script! My main script looks like the following:

For i = 1:1:1000
Building nodes and boundary conditions in the main script
source Material.tcl
source Section.tcl
Building Model (Elements, Loads, Mass) in the main Script
Recorders in the main script
Nonlinear Static in the main script using SuperLU
loadConst -time 0.0
Nonlinear Dynamic in the main script using Mumps

Then, according to your question! I would like to know whether my simulation will be faster if I split the main script into several files like what we usually do for Material.tcl? In such a case I can make several partitions. Thus, is it correct to say that if I partition my main script, then each partition will be handled by a processor?

Thanks for your help.
Best Regards,
A.H.Arshian

Post Reply