The reason for parallellazing lil-gp
is obvious and has already been mentioned earlier. But maybe it is useful
to introduce some term that are relevant in the field of parallel computing
in general. (I keep it brief I promise :)
speedup and efficiency: Speedup refers
the ratio of the computation time on a single CPU and on several CPU's.
The effeciency is the ratio of the speedup and the number of CPU's used
to achieve this speedup. What do these measures tell you. The speedup will
be bigger than 1 whenever the parallel implementation is faster the the
sequential one. (Note that the speedup is not necessarily >1) The efficiency
allows you the determine how well the used computational resources are
used. This can easily be understood with an extreme example. Everyone would
agree the a speedup of 3 using 100 CPU's would be a serious waste of computation
resources. Especially if you achieve a speedup of 2 using 20 CPU's for
Embarrasingly parallel: Some problems may
very well be hard to parallellize and thus exhibit some very low speedups
no matter the number of CPU being used. (This has been formally described
by Amdalh) At the other side of the spectrum of parallellisation one finds
the the embarrasingly parallel problems. These problem are by nature
parallel so that a parallel implemention can be done very efficiently.
Examples of that are : 1) cracking of keys, the group of all possible keys
can be divided and distributed over all the available CPU's. All the CPU's
can then work independently from each other. 2) Evoluationary computations
are also good canditates (well what do you know :). The idea of subpopulations
(or island model) implies the parallel evolution of different populations
with relatively infrequent exchanges between the populations.
PVM: Which stands for Parallel Virtual Machine
is a software package that provides both a API (C/C++ and Fortran) for
developing parallel programs and the environment to execute these
programs. PVM has been ported to several platforms (OS and hardware) including
massively parallel computers. PVM is freely available and can be
found very easily on each and every search engine on the web.
The information provided on this page is
copyrighted © by the
Free University of Brussels and provided as is.
From more information contact Johan Parent