FiatFlux-P - User-Level Workflow Design: A Bioinformatics Perspective - page 114

that is required for repeating the original FiatFlux-P workflow in the batch-
processing variant, this workflow can not be created completely by the current
synthesis algorithm, which is restricted to producing linear sequences of ser-
vices. It is, however, possible to predefine a sparse workflow model, in which
the looping behavior and other crucial parts are manually predefined, and
to subsequently fill in linear parts of the workflow automatically. Hence, the
synthesis problem is to create the loosely specified body of the FiatFlux-P
batch processing loop as shown Figure 5.10 (top).
The scan ranges and experimental data are read before the loop that it-
erates over the CDF data sets is entered (cf. the original batch-processing
FiatFlux-P workflow in Section 5.2). This is recognized by PROPHETS' data-
flow analysis mechanism, so that the synthesis is aware that services gener-
ating them do not have to be included in the solution. The reading of the
CDF input data and the storing of the results is however not yet contained
in the workflow. While the synthesis will recognize the necessity of including
the reading of input data from the interface specifications, the inclusion of
WriteFile is ensured by a domain constraint. Figure 5.10 (bottom) shows
one of the shortest solutions to this synthesis problem, where the CDF file is
read, the FiatFlux analysis steps performed, and the result finally stored in
a file.
By varying the constraints it is of course possible to obtain other solutions
to these synthesis problems, for example corresponding to the variations de-
scribed in Section 5.2. For instance, a constraint that enforces the use of
Netto CustomModel would not only make the synthesis include the respec-
tive SIB, but also the ReadNetworkModel SIB in order to have all required
input data available.
Again, the FiatFlux-P domain is comparatively small example, where
furthermore many general constraints are already contained in the domain
model. Hence the number of solutions is usually easily manageable and the
user only has to provide minor additional hints concerning his intentions
about the workflow in order to obtain the desired solutions.
With no constraints applied and searching for solutions up to length 7,
for instance, more than 5,000 solutions are found, while the domain con-
straints reduce the number of returned solutions to 44. Additionally en-
forcing the use of Netto CustomModel leaves mere 4 solutions. Note that
the (one) minimal solution for this synthesis problem is found in depth 6
and consists of ReadCdfFile , ConvertMS , ReadNetworkModelFile , Ratio ,
Netto CustomModel and WriteFile . The additional three solutions that are
found in depth 7 contain ConvertMS or Ratio twice or include the actually
not required ReadScanRanges .