A DATA GRID REPLICA MANAGEMENT SYSTEM WITH LOCAL AND GLOBAL MULTI-OBJECTIVE OPTIMIZATION
Loading...
Date
2009-05
Authors
ALMISTARIHI, HUSNI HAMAD E
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
As the scale and complexity of the scientific collaboration experiments grows, the
demand to an efficient and cost-effective data files sharing for solving large scale
problems is increased. Yet, providing efficient access to huge and widely distributed
data is still a considerable challenge and becoming a big problem in the domain. One of
the main solutions to the problem is that of data replication, which creates multiple
identical copies (replicas) of the same data file at different sites on the grid, and thus the
data availability and data reliability are enhanced. However, data replication may cause
increasing cost of the storage space. Thus, a good balancing of the number of replicas
and their appropriate locations in any replication system is required. However, the
current replication systems in grid environment are still few and lack of some important
issues such as Quality of Service (QoS), and thus they require more optimization
techniques.
In this thesis, we address the above problem by achieving two objectives namely, the
Local-Objective and the Global-Objective. The Local-Objective is a self-interest
objective for grid users' that aims at selecting the best required replica location from
among many replicas in minimum response time and high level of QoS. On the other
hand, the Global-Objective is a commonweal objective that aims at utilizing resources
xi
in steady state of the system m order to reduce storage space cost and network
bandwidth consumption.
Since the local objective has conflicting criteria measured by heterogeneous values, the
Analytical Hierarchy Process (AHP) model was used to solve the local objective
problem. Moreover, the local and global objectives may contradict to each other. Thus,
we proposed a replica management system that deploys: replication policy, replica
placement policy, replica replacement policy, and replica selection algorithm, in order
to enhance the replicas management in data grid. The advantages of the proposed
system are investigated in a simulation tool. The evaluation results demonstrated that
our system outperformed other existing systems in terms of: reducing the network
bandwidth consumption by 4.43%, reducing the storage space by 0.03 %, reducing the
job turnaround time by 10.30%, increasing fairness by 77.5%, and increasing the level
of QoS by 13 %. Therefore, we conclude that our proposed replication system can be
implemented in real data grid by providing the users' jobs with the required replicas in
reasonable time and quality. Grid users whom require data files and grid administrators
who manage grid resources can benefit from our system.
Description
Keywords
MANAGEMENT SYSTEM , GLOBAL