Database Reference
In-Depth Information
composition tools is to simplify for scientists the task of describing workflows.
Other,
task-based
systems such as Pegasus focus on the mapping and execu-
tion capabilities and leave the higher-level composition tasks to other tools.
Task-level workflow systems focus on resource-level functionality and fault-
tolerance, while service-level systems generally provide interfaces to certain
classes of services for management and composition. One important factor to
the adoption of workflow systems by scientists is the availability of workflow
tools and services that scientists can build on in order to create their appli-
cations. Such service availability forms part of the composition process since
it represents the available tools that can be composed within a system.
Pegasus takes a workflow description in a form of a directed acyclic graph
in XML format (DAX). The DAX can be generated using a Java API, any
scripting language, or using semantic technologies such as Wings.
35
In some
scientific applications, users prefer an interface that simply supports metadata
queries while hiding the details of how the underlying systems work. In astron-
omy, for example, users want simply to retrieve images of an area of the sky of
interest to them. In such cases Pegasus is usually integrated into a portal envi-
ronment, and the user is presented with a Web form for entering desired meta-
data attributes. Behind the portal a workflow instance is then generated au-
tomatically based on the user's input, given to Pegasus for mapping, and then
passed to DAGMan
37
for execution. Examples of this approach can be seen
in the Montage project (an astronomy application),
38
,
39
the Telescience por-
tal (a neuroscience application),
40
and the Earthworks portal (an earthquake
science application).
41
In all of these applications, Pegasus and DAGMan are
being used to run workflows on national infrastructure such as the TeraGrid.
Kepler provides a graphical user interface (GUI) for composing and edit-
ing workflows using a hierarchical representation of the workflow graph (see
the example in Figure 13.1). Dataflow is indicated by channels represented
as edges among the nodes of the graph, and each node represents either an
atomic task or a composite task (containing a subworkflow). The user interface
provides a semantic-search system across hundreds of different scientific com-
puting components available in the Kepler library. These components cover
a wide variety of scientific data processing and modeling activities, such as
geospatial data processing, signal processing, statistical algorithms, and data
transformations. The semantic search feature
42
,
43
assists the user in locating
components that are relevant to their analysis and modeling tasks. It is also
useful when searching the remote Kepler library, allowing users to find com-
ponents that have been shared by other Kepler users and to share their own
components and workflows with others. Kepler workflows can be executed
directly from the workflow-composition GUI or saved in an XML representa-
tion (MoML) and later passed to Kepler for execution in the absence of the
GUI. This feature allows Kepler to be embedded in Web portals and other
applications.
Taverna provides a GUI-based desktop application that uses semantic anno-
tations associated with services, employs the use of semantics-enabled helper
