Support for Cooperative Experiments in e-Science: From Scientific Work flows to Knowledge Sharing - Identification of Ligand Binding Site and Protein-Protein Interaction Area - page 141

7.4.2 Execution Support

Some of the characteristics that differentiate SWMS execution support from each
other and from other systems include the types of resources onto which the workflow
can be mapped and the method of enactment which describes the way workflow
tasks are scheduled and executed.
A common denominator amongst supported distributed resources is the Globus 2
grid middleware. SWMSs such as ASKALON, GWES, Karajan, Pegasus and Triana
readily support Globus due to its widespread use within the scientific community.
Some other SWMSs such as Pegasus and Triana are not only bound to one type of
resource but can interoperate with other middleware suites. A number of the studied
SWMS have the ability to use web/WSRF services as work fl ow components allowing
scientists to have access to a wide range of services which are publicly available
such as the BioCatalogue which currently hosts around 1,600 services. 3 Service-
oriented resources are exploited through SWMS such as Kepler, Taverna, Triana,
GridNexus and GWES.
Workflow enactment engines differ considerably between SWMSs though a
typical engine would consist of a scheduler which interprets the workflow graph
to deduce data/control flow dependencies and a runtime manager which takes care
of controlling individual tasks during runtime. ASKALON uses different schedulers
such as Heterogeneous Earliest Finish Time (HEFT), a genetic algorithm and a
myopic just-in-time algorithm. Similarly, GWES implements several schedulers.
A simple scheduler is the least-used scheduler, which chooses resources that had
not been used for the longest time. Another scheduler makes use of the Globus
Monitoring and Discovery System (MDS) so as to make a more informed deploy-
ment decision. Karajan supports late binding, hence deferring the decision of how a
task should be executed until the task is actually mapped to a resource. Pegasus uses
DAGMan as its scheduler. DAGMan is a directed acyclic graph meta-scheduler for
the Condor job scheduler. 4 Triana schedules groups of components where each
group can be piped or parallel. Early Taverna versions (1.x) used a heavily modified
version of the Freefluo 5 web-service orchestration engine while later versions (2.x)
replaced Freefluo for a new orchestration engine. In Kepler, directors represent
enactment engines for different MoCs.
Interoperability at runtime between SWMS is sometimes needed to take advan-
tage of the unique features of existing systems. Recent developments in the field and
the adoption of service-oriented architecture standards by most systems should
allow for interoperability. Triana takes this one step further by offering the possibility
to publish its (sub)workflows as a grid/web services. Kepler pursues another
approach which consists in wrapping the resources of other SWMSs, e.g., the
Nimrod engine to improve its parameter sweeping support (Abramson et al. 2009 ) .

2 Globus Middleware: http://www.globus.org/
3 http://www.biocatalogue.org/services
4 http://www.cs.wisc.edu/condor/dagman
5 FreeFluo: http://free fl uo.sourceforge.net