Migrating e-Science Applications to the Cloud: Methodology and Evaluation - Cloud Computing with e-Science Applications - page 93

Application Server

…

Service i OPAL

Database

Messaging
System

Workflow
Engine

Enterprise
Service Bus

Auditing

Database

Legend

Function call

Modeling &
Monitoring
Tool

Use

Message flow

Scientist

FIGURE 5.1
Main components of the SimTech SWfMS architecture.

web services through the service bus; that is, it sends a request for invocation
of a web service and receives the results back from the web services. The ser-
vice bus is also responsible for service discovery and selection if informa-
tion about concrete services to be used is not available during the workflow
deployment step.
The workflow engine also produces fine granular workflow execution
events and publishes them to the messaging system. These events are con-
sumed by the modeling and monitoring tool as well as by the auditing appli-
cation. The modeling and monitoring tool uses the execution information
to enable the live monitoring of running workflows. The auditing applica-
tion captures the same execution information and saves it into a database to
enable provenance and later analysis.
The actual workflow that serves as an example in the following is a kinetic
Monte Carlo simulation (KMS) that invokes several web services as part of the
simulation of solid bodies. These web services are implemented by modules
of the OPAL application (Sonntag et al., 2011). During their operation, the
OPAL web services access a MySQL database for both read and write opera-
tions. The example simulation of solid bodies is long running and requires
significant computing power. Speeding up the simulation was a challenge
that led to the decision to make use of cloud infrastructures, with the goal to
acquire additional computational resources and data storage for the time of
executing the simulation. This was indeed the major motivation for migrating
the simulation workflow system or parts of it to the cloud.
Since the example simulation produces big amounts of data, one of our deci-
sions was to temporarily migrate the database layer of the OPAL web services
into the cloud, thus realizing the migration scenario cloud bursting (Strauch
et al., 2013b), with Amazon Relational Database Service (RDS) as the migra-
tion target. However, migrating only the database layer to Amazon would