gives the possibility to perform parallel/distributed calculations without the
of a dedicated hardware but with a simple LAN PC network, and so permits to
drastically reduce computational time and memory requirements for large and
huge calculations without requiring huge hardware investments.
spite of what happens with trivial parallel solution strategies used by other
BEM and FEM packages, that simply assign to different processor different
frequency ranges, VNoise Cluster instead
use a real distributed strategy.
each frequency step the calculation is distributed across the network and all
the calculation steps are fully parallelized including Coefficients evaluation,
System solutions, and Microphone propagation.
strategy has the drastic advantage to not only reduce the computational time,
but also reduce the memory requirements. Infact each node keeps in memory only
its portion of the overall matrix, and therefore the in-core solver (that do
not perform disk swap) is limited by the sum of the overall RAM available on
all the connected PC and not by the single PC RAM.
permits not only to use different PCs for distributing large analysis, but also
permits to easily take full advantage of the great computing power of
recent multi-core CPUs. It is threfore possible to have an heterogeneous system
composed by several different PCs (single,dual quad core) all working together
in a completely transparent way from the user point of view.
high intrinsic scalability of BEM equations combined with the full parallel
architecture of VNoise Cluster, and with the
specifically optimized communication libraries permits VNoise cluster to
achieve a practically linear scalability and efficiencies.
The computational time decrease linearly with the number of computational
modes, and in some cases, due to the possibility to switch from out-core solver
to in-core solver, the computational time decrease even faster than the
number of CPU.
linear scalability remains unchanged in FEM for what concern the coefficient
evaluation stage, and it is only slightly reduced for the system solution
stage. Also in this case the memory is distributed across all the computer and
so it is possible to run multi-million nodes analyses with just a few PC.
Figure 1: The figure
above reports the scalability with the number of CPU or cores for a typical BEM
is based on proprietary, specific developed, communication library, and not on
MPI. The proprietary communication library is not only specifically optimized
to ensure fast communication, but also does not require any specific system
knowledge for installation and configuration.
The installation is as easy as installation of a standard windows
application on each computational node.
Also the execution of a parallel analysis is completely transparent from
the user point of view that has only to select the cluster solver, and then
start as usual the calculation.