Workflow Concept of WS-PGRADE/gUSE - Science Gateways for Distributed Computing Infrastructures - page 34

Fig. 3.1 Dataflow patterns supported by gUSE. a “ Task to Task ” pattern. b “ To Multiple Instance

Task ” pattern. c “ From Multiple Instance Task ” pattern. d “ Block Task to Sub-Workflow

Decomposition ” and “ Sub-Workflow to Block Task ” patterns. e “ Task Precondition — data value ”

pattern

The internal data interactions pattern class investigates data transfer and their
orchestration among the tasks. The simplest pattern supported by all of the data-
driven workflow management systems is called task to task pattern (Fig. 3.1 )
specifying that data can be passed between the tasks. According to how the control
and the data channels are used, three cases are de

ned: using data stored accessibly
globally from/to where the jobs can read/write; use integrated channels for control
and data transfer; or using distinct channels for sending control and data
information.
Integrated channels mean that the all the data generated by a job and the con-
trolling actions travel together to the next job. A disadvantage of this solution
appears in the case when part of the data set is not needed by the next job, but is
required by one of the subsequent jobs. In this case, the data come through the
intermediate job in vain. By contrast, transferring in distinct channels resolves this