The Trio_U code must fulfill the following requirements:
Transparency of the parallelism
Quality of the physical results: the same results must be obtained in parallel and sequential
To fulfill the point 2, the parallism method in Trio_U is based on a domain decomposition with overlapping.
Parallelism efficiency is assured by the use of standard libraries with explicit message passing (PVM or MPI), allowing to take the best benefit of the machine capacities (delivered by the manufacturer).
Parallelism transparency is reached through the encapsulation of the calls to these libraries and through the definition of a class for the tables of distributed values.