Database Reference
In-Depth Information
effective means are needed to discover data and software tools and to capture
workflow evolution.
68
Similarly, during workflow planning and execution, there
are numerous challenges, for example, how to eciently and reliably transfer
large amounts of data, or how to deal with distributed, heterogeneous system
environments.
Traditionally, in computer science, a core theme is optimization of program
runtime and memory usage by developing time- and space-e
cient algorithms
for the problems at hand. In many application areas, including scientific
workflows, “human cycles” are a sometimes neglected resource, which can
and should be optimized as well. For example, the use of data and work-
flow provenance information can be used for traditional purposes (such as
optimizing system performance or improving fault-tolerance
74
), but also to
enhance the scientist's insights when trying to understand or debug scien-
tific workflow results.
85
Similarly, approaches are needed to facilitate mod-
eling and design of scientific workflows that are easy to use. For example,
McPhillips et al.
22
list the following
user-oriented requirements
and provides
initial steps toward addressing them:
well-formedness
(facilitate the design
of well-formed and valid workflows),
clarity
(facilitate the creation of self-
explanatory workflows),
predictability
(make it easy to see what a work-
flow will do without running it),
recordability
(make it easy to see what a
workflow actually did do when it ran), and
reportability
(make it easy to
see if a workflow result makes sense scientifically). Clearly, the last two re-
quirements are related to capturing and managing provenance information,
a recurring theme in current scientific workflow research. Other research is-
sues mentioned in McPhillips et al.
22
are
reusability
(make it easy to de-
sign new workflows from existing ones) and
data modeling
(provide first-class
support for modeling scientific data), in addition to the already mentioned
optimization
issues (the system should take responsibility for optimizing
performance).
Acknowledgments
Work on Kepler is partially supported by the NSF (under grants IIS-
0630033, OCI-0722079, IIS-0612326, DBI-0533368) and the DOE (DE-FC02-
ER25809 and DE-FC02-07-ER25811); work on VisTrails is partially sup-
ported by the NSF (under grants IIS-0844546, IIS-0751152, IIS-0746500,
IIS-0513692, CCF-0401498, EIA-0323604, CNS-0514485, IIS-0534628, CNS-
0528201, OISE-0405402, IIS-0905385), the DOE SciDAC2 program (VACET),
and IBM Faculty Awards (2005, 2006, 2007, and 2008). Ewa Deelman's work
was funded by the NSF under Cooperative Agreement OCI-0438712 and
grant # CCF-0725332. We would like to thank Pierre Moualem, Meiyappan
Nagappan, and Ustun Yildiz for their support.
