Conclusion - User-Level Workflow Design: A Bioinformatics Perspective - page 190

is like oversimplifying the workflow composition problem by regarding it as
a jigsaw puzzle, where the goal is to find the one possible combination of
pieces. In realistic application domains for scientific workflow composition,
however, there are different pieces (services) that fit next to each other, and
consequently there are several possible combinations that are worth being
considered - also at the user level.
However, currently neither the basic jABC nor the PROPHETS manage
workflow variability explicitly, that is, workflow variants are more or less
created on the fly, rather than being systematically handled and described.
Consequently, it is desirable to incorporate explicit means for variability man-
agement [330] in order to support the development of variant-rich workflows
better.
As variant-rich workflows can be regarded as a particular form of software
product lines (SPLs) [253, 329], it suggests itself to pick up on the available
work on variability management of SPLs, such as [227, 99, 77, 65], here.
According to, e.g., [319, 169, 272, 143], two principal kinds of variability
modeling can be distinguished:
1. Structure-oriented approaches apply the concept of variation points [253,
Chapter 4], that is, explicitly defined points in the workflow model where
services or parameters have several alternatives. The different variations
that can realize the variation point are attached to it, and concrete work-
flows are built by selection of variants at the defined variation points.
2. Behavior-oriented approaches apply notions of labeled transition systems
[149] for representing behavioral variability (cf., e.g., [33, 71]). Concrete
workflow variants are derived by identifying the sub-systems of the model
that conform to the associated constraints or conditions.
Conveniently, both kinds of variability management are already supported
by the framework:
1. As detailed in [142, Section 4.1.4] and [143], the jABC's SIBs for building
hierarchical models can easily be interpreted as variation points in the
sense of structural variability modeling. The recently introduced Second-
Order Servification mechanism [239] goes further and makes it possible
to define variants of jABC SLGs via second-order parameterization, and
to exchange services and even subprocesses dynamically at runtime.
2. The constraint-driven workflow composition approach of PROPHETS
provides the means for a very liberal, declarative, and indirect way of
handling behavioral workflow variability [169, 271, 143]: Constraints in
terms of a domain model specify a set of similar workflows, that is, vari-
ants of an implicitly described, abstract workflow. The actual product
variants are automatically generated from these specifications and are
thus correct by construction [170].
Thus, adding explicit variability management support does in fact not
require comprehensive new developments, as several notions are already in-
herent in the present framework. It will rather be a question of consolidation