Degree Name

Doctor of Philosophy


School of Computing and Information Technology


Nowadays, the exponential data growth becomes one of the major challenges all over the world. It may cause a series of negative impacts such as network overloading, high system complexity, and inadequate data security, etc. Cloud computing is developed to construct a novel paradigm to alleviate massive data processing challenges with its on-demand services and distributed architecture. Data replication has been proposed to strategically distribute the data access load to multiple cloud data centres by creating multiple data copies at multiple cloud data centres. A replica-applied cloud environment not only achieves a decrease in response time, an increase in data availability, and more balanced resource load but also protects the cloud environment against the upcoming faults. The reactive fault tolerance strategy is also required to handle the faults when the faults already occurred. As a result, the data replication strategies should be aligned with the reactive fault tolerance strategies to achieve a complete management chain in the cloud environment.

In this thesis, a data replication and fault management framework is proposed to establish a decentralised overarching management to the cloud environment. Three data replication strategies are firstly proposed based on this framework. A replica creation strategy is proposed to reduce the total cost by jointly considering the data dependency and the access frequency in the replica creation decision making process. Besides, a cloud map oriented and cost efficiency driven replica creation strategy is proposed to achieve the optimal cost reduction per replica in the cloud environment. The local data relationship and the remote data relationship are further analysed by creating two novel data dependency types, Within-DataCentre Data Dependency and Between-DataCentre Data Dependency, according to the data location. Furthermore, a network performance based replica selection strategy is proposed to avoid potential network overloading problems and to increase the number of concurrent-running instances at the same time.

FoR codes (2020)

460601 Cloud computing, 460605 Distributed systems and algorithms, 460599 Data management and data science not elsewhere classified, 461299 Software engineering not elsewhere classified



Unless otherwise indicated, the views expressed in this thesis are those of the author and do not necessarily represent the views of the University of Wollongong.