Assembling DNA Fragments with a Distributed Genetic Algorithm - Parallel Computing for Bioinformatics and Computational Biology - page 293

Figure 12.2 Graphical representation of the parallel dGA.

Before moving on to Section 12.5, in which we analyze the effects of several
configurations of migration rates and frequencies, we introduce MALLBA which we
used in this work and where all our programs can be found.

12.4.2 MALLBA Project

The MALLBA research project [18] is aimed at developing a library of algorithms
for optimization that can deal with parallelism in a user-friendly and, at the same
time, efficient manner. Its three target environments are sequential, LAN, and WAN
computer platforms. All the algorithms described in the next section are implemented
as software skeletons (similar to the concept of software pattern) with a common
internal and public interface. This permits fast prototyping and transparent access to
parallel platforms.
MALLBA skeletons distinguish between the concrete problem to be solved and
the solver technique. Skeletons are generic templates to be instantiated by the user
with the features of the problem. All the knowledge related to the solver method (e.g.,
parallel considerations) and its interactions with the problem are implemented by the
skeleton and offered to the user. Skeletons are implemented by a set of required and
provided C

classes, which represent an abstraction of the entities participating in
the solver method:

++

• Provided Classes : They implement internal aspects of the skeleton in a problem-
independent way. The most important provided classes are Solver (the
algorithm) and SetUpParams (setup parameters).
• Required Classes : They specify information related to the problem. Each skele-
ton includes the Problem and Solution required classes, which encapsulate
the problem-dependent entities needed by the solver method. Depending on the
skeleton, other classes may be required.

Therefore, the user of a MALLBA skeleton only needs to implement the partic-
ular features related to the problem. This speeds considerably the creation of new
algorithms with minimum effort, especially if they are built up as combinations of
existing skeletons ( hybrids ).