There are two principal reasons for using parallel
rather than sequential PHOENICS.
to obtain computational results in a shorter time, and
to use finer grids than a single processor permits.
The parallel solver shares the computational domain and task between a number
processors; each processor then performs the computations for its part of the
domain simultaneously. Thus the whole task may be achieved
in a shorter time.
In theory, the more processors used, the shorter the computer time will be.
table shows the speed-up achieved when carrying out computations on a model with 20 million
It can be seen that there is not a linear relationship between the number of
processors and the speed-up ratio. This is because for each additional
processor added to the computation pool, there is an additional overhead.
Eventually a limit will be reached whereby adding additional processors does
not lead to a faster solution time. What this
limit is depends on the computational domain under consideration.
What is PHOENICS parallel
The EARTH module uses most of the computing time and therefore it is only
this part of PHOENICS, which is ported to the parallel computer.
The strategy employed is grid-partitioning (also called domain decomposition),
computational domain is divided into sub-domains, each of which is assigned
to a particular processor having its own identical copy of the EARTH executable.
During the computation, each processor exchanges information with its
neighbours at the appropriate times.
In the parallel-EARTH implementation, a single processor controls the
acting as a server to the other processes; this is called the 'Master'
Its main input operation is the reading from disk of the problem-defining
data files, and the broadcasting of the appropriate information to the
other processes. In parallel, each process extracts information specific
to the sub-domain it belongs to and continues with the solution procedure.
At the end of computations, the results from each processor are sent to the
Master, there to be
which are re-assembled and written on the disk as a single PHI file, exactly
as in a sequential run; an option for writing individual PHI files also exists.
A single RESULT
file is also written by the Master processer; there is also an option for
writing individual RESULT files).