Lessons Learned - User-Level Workflow Design: A Bioinformatics Perspective - page 144

Fig. 7.6 Synthesis statistics (visited nodes) for the GeneFisher-P scenario
which can be about 7 times faster (more precisely: about 30 seconds instead
of roughly 3.5 minutes on the machine on which the experiments were car-
ried out). As another example, the synthesis expands 3,642,130,982 nodes
in depth 5 when applying constraints 1 and 2, while the application of all
constraints leads to 327,212,871 visited nodes, which is still about 6 times
faster (about 14 minutes instead of 82). However, the constraints reduce the
number of visited nodes not always as clearly as they restrain the number
of solutions. For example, the number of visited nodes in depth 5 for the re-
finements 1-3 are the same order of magnitude (roughly 3.5 billion expanded
nodes), whereas the corresponding numbers of solutions are clearly different
(cf. Figure 7.1).
The EMBOSS domain model is indeed a prominent example for a large
collection of similar services that induce a high branching factor in the syn-
thesis universe and accordingly a large solution space. Consequently, the
search depths that can be reached in acceptable synthesis execution times
are comparatively small unless extremely concise constraints are provided.
In contrast, the GeneFisher-P and FiatFlux-P domain models comprise con-
siderably less and furthermore behaviorally distinct services, and hence the
corresponding solution spaces are smaller. As Figures 7.6 and 7.7 show, there
greater search depths can more easily be reached, as the number of visited
nodes “explodes” much later.
Figure 7.6 shows that the number of visited nodes in the GeneFisher-
P scenario is more or less equal and comparatively small in the beginning,
regardless of the actual constraints that are applied. In a search depth of
10, the difference becomes more striking, as 60,929,232 nodes are visited
in the unconstrained case, whereas the three constrained cases cause the
synthesis algorithm to expand between 1,165,825 and 9,091,818 nodes. This
took between 2.8 and 17.5 seconds instead of 1.5 minutes. In depth 12, the