Download files
Abstract
Rapidly growing Internet-based services have substantially redefined the way of providing data persistence and retrieval from that of the one-sizefits-all solution offered by relational database management systems to a full spectrum of cloud databases solutions. This significant paradigm shift did not happen spontaneously, but its progress and adoption was hastened by the boom in cloud computing adoption. Cloud computing also represents a new resource provisioning paradigm that shifts the location of resources to the network to reduce the costs asso ciated with the management of hardware and software resources.
This thesis takes the unique cloud platform customer's perspective and explores in detail the trade-off characteristics between performance gain and monetary cost across different cloud platforms. These related problems are studied: 1) generic performance evaluations of different cloud providers; 2) the service level agreement (SLA) gaps between the cloud providers and the cloud customers.
The design and architecture of cloud varies among cloud providers. For performance evaluations, this thesis spends the initial two chapters on addressing two generic evaluation solutions for different cloud platforms and cloud databases respectively. The first solution proposes a generic evaluation framework that focuses on performance, availability, and reliability characteristics of various cloud platforms. The second solution provides a generic benchmark architecture for benchmarking cloud databases, specifically NoSQL database as a service. It measures the performance of replication delay and monetary cost.
As existing SLAs of cloud providers guarantee only the availability of their services, rather than supporting the straightforward requirements and restrictions under which SLAs of cloud customers' applications need to be handled, this thesis uses another two c hapters to further investigate the approach for the customer's interest to automate SLA-driven management for database replication on virtualized database servers. The investigation takes two steps. In the first step, the performance of database replicatio n of virtualized database servers are comprehensively evaluated. The second step takes the lesson learned from the first step to build a SLA-driven framework for managing database replication. The framework implements customer-centric dynamic provisioning mechanisms for virtualized database servers based on adaptive application requirements.