The OpenSees Community

Hi,

I am conducting a pushover analysis of a large model using OpenSeesSP. Model runs fine with np = 1 but takes about 10 hours to complete. When I switch to np > 1, gravity analysis completes successfully but then nothing happens afterwards; no errors messages and no sign of completion even after 10 hours. I am using mumps and displacement control and have rigid diaphragm in the model and hence using Transformation constraint handler. One more thing is that in the output folder I get files with .1 and .2 extensions depending on how many processors I specify. I could see some of gravity analysis results in the output file I specified in recorders but no results from any converged pushover analysis (nothing in .1 and .2 files though). I am running model using mpiexec command. Any thoughts on what might be happening?

Thanks.

can you try removing the static analysis and just doing a few steps of the transient to see if that works.

Hi fmk,

The same story with transient analysis as well.

Thanks.

Any thoughts fmk? Thanks.

sorry .. missed your reply .. you did remove the gravity analysis and that failed to solve the problem? .. if you did and it didn't i will need to get the code or at least look at it .. are you using NEEShub or is this your own machine . .if NEEShub you can share a session with fmk.

Hi fmk, sorry, I was doing a transient analysis following gravity. It appears to be working when I only do transient. What does this mean? Thanks.

if the first analysis is working and not the second the problem lies in the Mumps solver (it seems to happen a lot (probably it happens occasionally i only hear about it when there are problems with it)) .. one possible solution is to do a gravity analysis with a different solver and then use Mumps for the time consuming calculations.

Thanks fmk, that sounds like a good idea but for some reason I cannot use more than 1 processors for other than Mumps. I would like to use Mumps for the for time consuming analysis and some other solver for gravity analysis. Is it possible to do gravity analysis with np = 1 and then do more time consuming analysis with np more than 1?

no.
so mumps fails with np=1 when you do static and then transient. if so i would love to see the script as it might help me see if i can't track down this problem ith mumps.

Thanks fmk. I will shortly send you the script.

Hi fmk, to answer your questions:

1. mumps does Not fail with np = 1 when I do static and then transient. The problem starts only for np > 1.
2. With np > 1, I can only do a transient analysis but not the transient following a static analysis.

Would you still like to see my script?
Thanks.