Workflow Concept of WS-PGRADE/gUSE - Science Gateways for Distributed Computing Infrastructures - page 33

gUSE applies an XML-based language for de ning workflows and their

graphical representation, including the structural con guration as well as execution

information. Thus, the XML description of a gUSE workflow consists of three parts
de

nes the
workflow as a set of nodes with associated ports including x and y coordinates for
the canvas of the Graph Editor. The

ned as tags: “ graph ” , “ real ” and “ instances ” . The “ graph ” tag de

guration infor-
mation for the jobs and for the ports as well, extended by a history record that keeps
track of changes in arguments. The

“ real ” tag contains all con

tag stores reference records and
brief status information about the executed workflow instances. At design time, the
Graph Editor (introduced in Chap. 2 ) and WS-PGRADE UI ease the creation and
the editing of workflow descriptions for the user.

“ instances ”

3.3 Work fl ow Patterns

The Workflow Patterns Initiative (workflowpatterns.com) keeps track of scienti c

papers citing workflow pattern-related work (until 2009 but there are many more)
showing that workflow patterns became a widely accepted way for designing and
for re-factoring workflow applications. Taverna is used for investigating solutions
for parallelism and pipeline processing (Missier 2010); others such as Yet Another
Workflow Language (YAWL) (Van Der Aalst 2005) or (McPhillips 2009) are
speci

ed directly using workflow patterns. Patterns play roles in (Yu 2005) to set up
taxonomy for workflow structures. Since gUSE uses a data-driven workflow lan-
guage, we focus on the identi ed dataflow patterns only.

3.3.1 Data fl ow Patterns

Various pattern classes are identi ed in (Russel 2007) for data-driven workflow

management systems such as visibility of data, internal or external data interaction
or data-based routing . Several patterns are introduced clearly and are detailed to
show the different cases within a class. For instance, the data visibility class con-
tains patterns describing different scopes for accessing the data from the tightest
task level until the widest environment level. The data interaction class incorporates
the possible ways to communicate between tasks, focusing on their additional
properties, e.g., communication with a task that represents multiple instances, or a
subworkflow decomposition. In the followings we introduce those patterns that are
supported by WS-PGRADE/gUSE system. Such patterns that are supported
explicitly are shown in Fig. 3.1 .
In the class of data visibility, the task data pattern is supported by gUSE lan-
guage only. Task data interprets the strictest possibility; it allows accessing data
only for the job similarly to the de nition of private variables in the object-oriented

programming languages.