Information Technology Reference
In-Depth Information
5.1 Introduction
With the advent of grid technologies, scientists and engineers are build-
ing complex and sophisticated applications to manage and process large
datasets and execute scientii c experiments on distributed grid resources
[1]. Building complex workl ows requires means for composing and exe-
cuting distributed applications. A workl ow expresses an automation of
procedures wherein i les and data are passed between procedures appli-
cations, according to a dei ned set of rules, to achieve an overall goal [2]. A
workl ow management system dei nes, manages, and executes workl ows
on computing resources. The use of the workl ow paradigm for application
composition on grids offers several advantages [3] such as:
•
Ability to build dynamic applications and orchestrate the use of
distributed resources
Utilization of resources that are located in a suitable domain to
•
increase throughput or reduce execution costs
Execution of spanning multiple administrative domains to obtain
•
specii c processing capabilities
Integration of multiple teams involved in managing different parts
•
of the experiment workl ow—thus promoting interorganizational
collaborations
Executing a grid workl ow application is a complex endeavor. Workl ow
tasks are expected to be executed on heterogeneous resources that may be
geographically distributed. Different resources may be involved in the
execution of one workl ow. For example, in a scientii c experiment, one
needs to acquire data from an instrument, and analyze it on resources
owned by other organizations, in sequence or in parallel with other tasks.
Therefore, the discovery and selection of resources for executing work-
l ow tasks could be quite complicated. In addition, a large number of tasks
may be required to be executed and monitored in parallel and the location
of intermediate data may be known only at runtime.
This chapter presents a workl ow enactment engine developed as part of
the Gridbus Project at the University of Melbourne, Australia [4]. It utilizes
tuple spaces to provide an event-driven mechanism for workl ow execu-
tion entities. The benei ts of this design include the ease of deployment for
various strategies of resource selection and allocation, and supporting
complex control and data dependencies of tasks with scientii c workl ows.
Search WWH ::
Custom Search