Workflow Concept of WS-PGRADE/gUSE - Science Gateways for Distributed Computing Infrastructures - page 45

3.7 Pattern Compositions

Based on the previously introduced implementations of data patterns, in gUSE
more complex control patterns can be created by composing and specializing them.
Control patterns are identi

ed by Van Der Aalst (2005), and they investigate several
commonly used open-source and commercial workflow management systems. In
the following we introduce composition of control flow patterns and their imple-
mentation in gUSE using the features designed for data patterns.

3.7.1 Parameter Sweep Sequence

One of the simplest compositions of control patterns is the commonly used and
widely known parameter sweep (PS) sequence. It is an ideal structure if large
separable and therefore parallelizable parameter fields must be elaborated by the

same application. It can be designed as a composition of “ multiple instances with a

priori run-time knowledge ” pattern, which speci es the job generalization

according to the number of data sets, and the “ general synchronizing merge ”

control pattern that de nes a structure for collecting the results of the parallel

execution. Using the generator and collector port properties, we can de ne the

frequently used PS workflow pattern consisting of a generator node, a PS node, and
a collector node (Fig. 3.6 ). The role of the generator node is to split the large input
data

files and pass
them to the PS node via its generator port (output port 0). The workflow interpreter
will generate from the PS node as many jobs (PS node instances) as there are small
data

file received in input port 0 into a number N of small size data

files ( N ) generated by the generator node. All these N PS node instances
generate one output file. These N output

files are gathered via the collector input

Fig. 3.6 PS workflow pattern consisting of a generator node, a PS node, and a collector node