Page 297 - The Combined Finite-Discrete Element Method

280 COMPUTATIONAL ASPECTS

to, say, the computer being accidentally turned off. In such cases, it is customary to save
whole in-core database say 10 or 20 times during the run time. In this way, the maximum
loss cannot exceed 10% or 5% of the total required run time. The job that has crushed
once will therefore require only 10% or 5% longer run time than it would take if the
crush did not occur.

9.1.5 Maximising transparency

Large scale combined finite-discrete element simulations are so diverse in applications
that two different applications may not even look alike. The algorithms employed in
application A may completely differ from algorithms employed in application B. Both
applications may share the basic common concepts, such as contact detection, contact
interaction, solver and discretisation. A general purpose combined finite-discrete element
computer program should in theory handle a wide range of applications. This can be done
in two ways:

• Implement common algorithms in such a way that they fit all applications. However,
large scale combined finite-discrete element simulations are very demanding on com-
puter resources, and generalised implementation of common algorithmic procedures will
neither minimise RAM nor CPU requirements. In fact, such an approach is more likely
to maximise both. Thus the classic approach of ‘all tools in one computer program’ is
not an option in large scale combined finite-discrete element simulations.
• Produce a virtual workbench-like or library-like software environment, which would
enable a relatively fast and automated process of building a specialised computer pro-
gram for a given application, or even for a given problem.

The end goals of reusability, transparency and CPU and RAM efficiency seem to contra-
dict each other. Indeed, to achieve both in languages such as FORTRAN is impossible.
A bold move towards one of the modern day languages is necessary. In terms of speed,
transparency and speed of implementation, it is now widely recognised that modern lan-
guages such as C or C++ offer unique advantages in comparison to languages such as
FORTRAN (in terms of transparency) and in comparison to languages such as Java in
terms of speed.

9.2 VERY LARGE SCALE COMBINED FINITE-DISCRETE
ELEMENT SIMULATIONS

Very often practical application of both discrete elements and the combined finite-discrete
element method require systems comprising hundreds of millions of discrete elements.
These simulations are termed very large scale combined finite-discrete element simula-
tions. RAM requirements for such systems are at the upper limit of the most present
day affordable single processor machines. The problems with RAM requirement are best
illustrated by the fact that these systems may require tens, even hundreds of gigabytes
of RAM space. With the virtual memory concept, this space must be addresses. Virtual
memory can be viewed as an array with the first variable being at address 1, second