Adaptive resource optimization of three-tier web applications running on the cloud

Cloud Computing has been envisioned as a promising approach and dominant computing model in IT infrastructures. It employs the Virtual Machine (VM) technology to provide on-demand provisioning of resources on a pay-as-you-go base. This motivates enterprise application providers to adopt Cloud computing and outsource their infrastructures and computational needs. In particular, Cloud computing has become an attractive and promising platform for three-tier web applications. However, an inappropriate and inefficient resource management practice may negatively affect the Service Level Agreement (SLA) and the response time experienced by users, essentially under high load operating conditions. Furthermore, this may result in substantial amount of energy consumption in data centers, which consequently leads to a high operational cost. This research consists of three major parts. The first part is a benchmark study that runs a three-tier web application on public cloud providers. It proposes WPress benchmark, which is based on the widespread blogging software, WordPress, as a three-tier web application, and implements an open source workload generator. Furthermore, a CPU micro-benchmark is utilized to investigate CPU performance of cloud-based VMs in greater detail. The main objective of this study is to evaluate and compare the average response time and operational cost of three-tier web applications running on commercial cloud providers. The small and large instance types of Amazon EC2, Microsoft Azure, and Rackspace Cloud are evaluated. Based on the experimental results it is found that Rackspace and Microsoft Azure are the preferred cloud solutions for small and large instance types, respectively. Furthermore, it is noticed that average response time has substantial fluctuations for large instances that can lead to significant SLA Violation (SLAV) for high load. Cloud service providers are penalized if the service level agreement is violated. This may result in the loss of revenue. It is hence important for service providers to minimize SLAV as much as possible. To address this, the second part of the thesis introduces a network-aware VM migration algorithm to minimize the average SLAV of the system. The algorithm considers steady-state traffic condition to minimize the negative effect of migration on other flows. The Network Gain (NG) is calculated for candidate VMs and the VM with the maximum NG is selected. The simulation results in CloudSim show that the suggested algorithm yields significant performance improvements. The increase in computing capacity and communication units in modern data centers results in the high energy consumption and operational cost. This negatively affects the environment and the cost of using cloud. Therefore, it is crucial to design new approaches for saving the energy consumption in the data centers. The third part of this thesis is to address this issue by extending the network-aware algorithm with energy-awareness capabilities to minimize the energy consumption while maintaining the SLA. In addition to NG, Power Gain (PG) is calculated for each candidate VM and two lists are created for each congested link: NG list and PG list. The VM with the lowest sum of the rank is selected. An extensive simulation study in CloudSim presents the effectiveness of the proposed approach.Doctor of Philosoph