Abstract:
E-science applications require discovering, collecting, transferring and processing large
volumes of scientific data. In divisible load e-science applications, data is generated
and stored in geographically distributed repositories (e.g., instruments, sensors, cam-
eras, satellites and storage facilities). The generated data can be divided into in-
dependent subsets to be analysed distributed at many computing locations. Such
applications usually require optical networking for fast and reliable data transfer. In
this thesis, we propose a framework for divisible load applications in optical grids.
Within this framework, schedulers are developed to co-schedule computational and
optical network resources. Moreover, processes to handle divisible load applications
in different architectures including centralized, hierarchical, peer-to-peer and super-
peer are proposed. Fault management techniques are introduced to handle network
faults by considering distinctive characteristics of e-science applications and optical
grids. This research conducts a comprehensive study of different algorithms, tech-
niques and processes that have been introduced within the framework. The results
provide guidelines to build more efficient, scalable and reliable optical grid systems.
The results can serve as a guide to the best choices in selecting the scheduling algo-
rithms, fault management techniques and architectures according to different network
and application parameters.
Description:
A Thesis Submitted to the Faculty of Graduate Studies and Research in Partial Fulfillment of the Requirements for the degree of Doctor of Philosophy in Engineering, University of Regina. xiii, 246 l.