Abaqus is a general purpose finite element analysis package. It can
be used to simulate the response of structures and solid bodies
to loads, impacts and thermal stress and to visualize the results
of simulations.
Abaqus is available to all NCSA
academic users on the following systems:
- EMT64/Linux (abe.ncsa.uiuc.edu), version 6.7-1
- SGI-Altix/Linux (co.ncsa.uiuc.edu), version 6.7-1.
Commercial use of Abaqus is governed by the NCSA Private
Sector Program. Please contact Seid Koric at skoric@ncsa.uiuc.edu
for more information.
The graph on the left side shows Abaqus scaling performance
on NCSA's Origin 2000 for various parallel runs of 210,000 and
17,000 dofs fixed problem sizes using a built-in Abaqus procedure
for rectangular beams under torsion loads.
Speedup (Sp) is defined as the wall clock time for a single processor
run divided by the wall clock time using n# processors. It tells
us how much performance is achieved by using multiple processors,
or the benefit of solving the problem in parallel.
It can be seen that fairly good Speedup is achieved for the 210K dofs
problem. The 17K dofs curve shows very limited Speedup even with
2 processors, while for
16 processors a point of negative return is reached due to too much
parallel overhead compared to the amount of computation. Speedup
is limited when the problem size is too small. Therefore please
don't use more than 4 processors unless your job has more than
200,000 dofs.
Speedup is always an increasing function of problem size and the more
dofs one has, the more processors should be used to achieve optimum
performance. Speedup is not a function of the size of time-dependant
Abaqus procedures, and it completely depends on the number of dofs
in your FE mesh domain.
The graph on the right side is showing Abaqus wall clock data for
1 million dof problem ran on SGI O2K, IBMp690, and HP SuperDome.
