Database Reference
In-Depth Information
after failure, this actor checks the current tasks to be performed against the set
of successful tasks and skips over any that were already executed successfully.
In this way, the next actor in the pipeline can immediately start working on
it (or skip over it as well). Thus, although the workflow restarts from the very
beginning, pushing the initial input tokens back into the pipeline, the actors
“fast-forward” to the most recent state prior to the workflow failure, by skip-
ping the tokens corresponding to previously successful tasks. The time spent
by the workflow engine in this fast-forward restoration process is negligible
compared to the time of actually executing the remote operations.
13.4 Grid Workflows and the Scientific
Workflow Life Cycle
The term
grid workflow
applies to workflows that employ distributed (of-
ten wide-area) computational resources (often referred to as “the grid”). Like
other scientific workflows, grid workflows can be seen as high-level specifica-
tions of sets of tasks and the dependencies between them that must be sat-
isfied in order to accomplish a specific goal. The specific goal of grid-enabled
workflow systems is to reduce the programming effort required of scientists
orchestrating a computational science experiment in a wide-area, distributed
system. The vast majority of scientists do not use grid systems in their day-
to-day practices, largely because of usability barriers. Workflow systems are
beginning to address these usability barriers and to make grid computing
far more accessible to general science users. In this section, we focus on the
first three stages of the life cycle summarized in Section 13.2.1 — scientific
workflow composition, mapping of workflows onto resources, and workflow
execution — and describe how several popular grid-enabled workflow systems
support these different stages. We present a cross-sectional view of the types
of grid workflows that are currently being deployed and compare features pro-
vided by Kepler,
18
,
28
Pegasus,
29
,
30
Taverna,
14
,
31
,
32
Triana,
33
,
34
and Wings.
35
,
*
13.4.1 Workflow Design and Composition
Most e-Science workflow systems provide a graphical tool for composing work-
flows. For example, Kepler and Triana have sophisticated graphical compo-
sition tools for building workflows graphs using a graph or block diagram
metaphor, where nodes in the workflow graph represent tasks, and edges rep-
resent dataflow dependencies or task precedence. The intent of these graphical
*
For a high-level overview and attempt at a classification of current systems see
36
and
http://www.extreme.indiana.edu/swf-survey/; these include references to other scientific work-
flow systems, such as Askalon, Karajan, and many others.
