Related Work - User-Level Workflow Design: A Bioinformatics Perspective - page 181

8.2.4 Summary

Taverna's focus on easy, automatic import of common bioinformatics services,
in combination with its data-flow-based workflow models, makes it convenient
for rapid development of batch processing jobs and computationally simple
workflows. Modeling the data-flow, the workflow developer takes a resource-
oriented perspective, which is close at the actual data that is processed by the
workflow. More complex control-flow constructs, like conditional branching
or looping, can however be dicult or impossible to realize.
In contrast, Bio-jETI emphasizes the computational power of its workflow
models and provides thorough control-flow structures, which are necessary for
more complex analysis problems. The control-flow-oriented approach further-
more allows workflow developers to think in “Dos” and “Don'ts” and steps
and sequences of action in their own terms at their level of domain knowledge.
The data-flow is modeled separately in Bio-jETI following a shared-memory
approach. In case of computationally simple workflows, this leads to the prob-
lem that some users feel they were doing the same work twice, namely when
they draw branches between SIBs and have to configure their parameters so
that the data is passed from the one SIB to its successor. Readily configured
linear (sub-) workflows can, however, be automatically generated by using
the synthesis feature of the PROPHETS plugin given that an appropriate
domain model is available.
Naturally, the choice for a particular workflow system always depends on
the concrete use case and to some extent also on the skills and preferences
of the workflow developer. As discussed above, however, the availability of
control flow structures and flexible data exchange facilities (such as a shared
memory) is necessary when it comes to analysisproblemswith certain compu-
tational complexity. In fact, it has been pointed out already in [336] that most
“workflow systems in this domain lack the facilities to model advanced con-
trol structures such as conditional branches and iteration”, and that “there
is growing demand for a more controlled approach to workflows in the life sci-
ence domain”. Hence, systems like Bio-jETI, which make use of control-flow
workflow models, are generally preferable.