Information Technology Reference
In-Depth Information
decentralized execution of composite BPEL services [67-70]. They
have investigated how to partition a BPEL program into multiple parts,
especially the partitioning of fault-handling code; they model partition
policies to improve execution performance; and they consider how to
partition the model when dataflow is constrained. Recently, a process-
mining-based model fragmentation technique has been proposed for
distributed workflow execution [71].
1.3.5 Scientific Workflow Systems
Besides the business community, the scientific community has shown
growing interest in workflow technology and has exploited its power in
Grid computing, scientific investigation, and job flow management
[72,73]. Essentially, a scientific workflow is a specialized workflow
orchestrating computation and data manipulation tasks into a process of
scientific value. A scientific workflow system becomes prominent with
the arising interest in data-intensive science,ore-Science [74]. In
e-Science, scientists are facing an enormous increase in raw data
from various resources, such as telescopes, instruments, sensor net-
works, accelerators, and supercomputers. For example, in high-energy
physics, the main detectors at the Large Hadron Collider (LHC)
produced 13 petabytes of data [75] in 2010. In bioinformatics, 1330
molecular biology databases [76] were reported in 2011. Among them,
GenBank, the US NIH DNA sequence database, contains more than 286
billion entries for more than 380,000 organisms [77]. To conduct any
nontrivial analysis using large amounts of data, scientists need the help
of a workflow system.
There are many scientific workflow systems available and the edited
topic [72] provides a good summary of them. Each of them provides a
graph-based interface for service composition, with an underlying work-
flow metamodel. The workflow metamodels used by these service-based
systems are either adopted from industry standard or homegrown.
GPEL and OMII-UK have adopted BPEL. Adopting BPEL can
bring advantages such as rigorously defined model syntax and seman-
tics, readily available software tools, and portability of workflow
specifications. However, scientific workflows have a particular focus
on data flow (versus control flow in business workflows) and parallelism
(versus the complex logic in business workflows), and are tightly
integrated with the underlying computation infrastructure. To deal
