Information Technology Reference
In-Depth Information
The researchers that plan the experiments are experts of a certain domain,
not skilled programmers. Thus, a helpful workflow management system for
scientific research must support intuitive and rapid process design, ideally on
a level of abstraction that focuses on defining essential aspects of the workflow
(such as the execution order of services, input data, and transition conditions
[166]) rather than on technical and syntactical details of a (standard) pro-
gramming language. This implies a software architecture that separates the
engine that deals with the heterogeneous services from the interface that
presents a unified world to the end user, ideally by means of a comprehensive
and intuitive graphical user interface that makes the system easy to use and
facilitates agile workflow development. Graphical workflow representations
are furthermore advantageous as they directly provide intuitive documenta-
tion of the computational experiment [128].
Requirement 2: powerful workflow model
A powerful workflow model for scientific applications, that is, a model which is
fit for implementing complex analysis procedures, has to comprise functional-
ity for control-flow handling, data handling, and hierarchical modeling [128].
•
Requirement 2.1: control-flow handling
The ability to define basic aspects like the order of workflow steps and
transition conditions between them is of course essential, ideally also syn-
chronization requirements should be expressible [166]. Furthermore, con-
trol flow structures like conditional branching and conditional loops are
required when building complex workflows. As discussed in [128], pure
“for each element”-loops, which allow for iteration over the elements of a
list, are in general not sucient.
•
Requirement 2.2: data handling
Typically realizing data analysis processes, scientific workflows rely on
input data from which they finally derive (a set of) results, usually also
producing a number of intermediate or partial results [166]. Hence, sci-
entific workflow systems have to provide adequate support for defining
the flow of data between the services in the workflow. Technically, there
are basically two possibilities for passing data between the single services
within the workflow: transferring the data via “pipelines” from one service
to another (messaging approach) or assigning identifiers to all data items
so that the services access them via named variables (shared-memory
approach).
As for the supported data items themselves, it is mandatory to support
primitive data types and (one-dimensional) lists [128], but also complex
data types and multi-dimensional lists are frequently needed. In order
to make it easy for the user to utilize the data handling capabilities of
the workflow model, workflow systems should furthermore support basic
data processing functionality (such as arithmetic operations, string ma-
nipulations and sub-data access methods [128], but also means for data
Search WWH ::
Custom Search