In the presented approach, the parallelization strategy is based again
on the domain decomposition concept. The domain decomposition,
however, has been considered on two levels - the model level and
the model entity parametric tree level. The decomposition on model
level splits the model into domains on the model entity basis.
Each model entity which does not form boundary of another model entity
of higher dimension (these entities are called the relevant model
entities thereafter) constitutes an individual domain. The
advantage of this approach consists in the fact that the decomposition
is in agreement with the physical concept of the model (the model
boundaries coincide with the interprocessor boundaries).
The decomposition on the parametric tree level may be compared, in
some sense, with the ORB in the parametric space of a given model
entity. Since the splitting is done in the parametric space its image
in the real space does not generally result in domains of similar
size. The next bisection, however, may be applied to larger (or more
complex) from those domains etc. In this way, an arbitrary number of
domains, with no one exceeding a prescribed value of complexity, may
be created. This is a good starting point to achieve acceptable
parallel performance for various number of processors and varying
model complexity. An important fact is that all the domains can be
handled uniquely as model entity domain because they are described by
the same model entity and differs only in the range of parametric
coordinates.

*Daniel Rypl
2005-12-03*