Database Reference
In-Depth Information
of workflows — these collections are described as Packs or Research Objects.
Hence, rather than a workflow repository, myExperiment can be seen as an
aggregator and registry of scientific objects — as its name suggests, it deals
in experiments.
13.7 Conclusions and Future Work
Scientific workflows are increasingly being adopted across many natural sci-
ence and engineering disciplines, spanning all conceivable dimensions and
scales, from particle physics and computational chemistry simulations, to
bioinformatics analyses, medicine, environmental sciences, engineering, ge-
ology, phylogeny, all the way to astronomy and cosmology, to name a few.
Not surprisingly, with these rather different domains come different require-
ments for scientific workflow systems. While some scientific workflows can
be conveniently executed on a scientist's laptop or desktop computer, others
require significant computational resources, such as compute clusters, possi-
bly distributed over a local or wide area network. In this chapter, we have
given an overview of common features of scientific workflows, described the
phases of the scientific workflow lifecycle, and provided some background on
the different computational models (and other differences) of scientific work-
flow systems. A detailed case study from plasma fusion simulation was used to
take a closer look at the challenges when managing simulation workflows. We
also provided an overview of some of the different approaches taken for scien-
tific workflow systems, focusing on workflow composition, resource mapping,
and execution. Furthermore, we described approaches for runtime monitor-
ing and provenance management in scientific workflows, and finally discussed
workflow sharing and reuse using a “Web 2.0 approach.”
The area of scientific workflows is dynamic and growing, as evidenced
by many workshops, conferences, and special issues of journals, devoted to
the topic (e.g., see Taylor et al.,
6
Gil et al.,
8
and Ludascher and Goble,
82
and Fox and Gannon
83
). Numerous challenges of scientific workflows remain
and require future research and development. For example, findings from an
NSF-sponsored workshop on scientific workflow challenges are reported in Gil
et al.,
8
where they are grouped into
application requirements
(e.g., supporting
collaborations, reproducibility of scientific analyses, and flexible system en-
vironments),
sharing workflow descriptions
(e.g., how to represent and share
different levels of workflow abstractions),
dynamic workflows
(how to support
the exploratory and dynamic nature of scientific analyses), and
system-level
workflow management
(e.g., how to scale workflows and how to deal with
infrastructure constraints). In, Deelman and Chervenak,
84
data-management
challenges of data-intensive workflows are presented and organized according
to the data lifecycle in a workflow. For example, during workflow creation,
