Virtualization enables the ability to consolidate multiple servers on a single physical machine, increasing the infrastructure utilization. Maximizing the ratio of server virtual machines (VMs) to physical machines, namely the consolidation ratio, becomes an important goal toward infrastructure cost saving in a cloud. However, the consolidation can also cause performance degradation, jeopardizing Service Level Agreement (SLA). To maintain the SLAs, previous work builds control systems on top of existing resource schedulers to tune the resource allocations based on the usage and VM performance. Without controlling both the resource allocations and the resource access order, it is difficult to effectively control the response time. The existing schedulers normally let the VMs use the resources proportionally to their resource allocations in a round-robin manner. The response time becomes worse as the number of VMs increases. This approach also performs the allocation re-adjustment based on the resource usage observation periodically taken every multiple seconds. This is not sufficient to satisfy the fine-grained target response time where hundreds of milliseconds can affect web service revenues. To support the fine-grained SLAs, we propose VM resource scheduling called "Pack[superscript2]" which consists of a CPU scheduler called "CPack" and a disk scheduler called "DPack." Both "CPack" and "DPack" integrate the SLA requirements and the VM performance into the scheduling decisions. This allows "CPack" and "DPack" to quickly adjust both the resource access order and the resource allocations to avoid the SLA violation. The schedulers essentially schedule the VMs that are more likely to fail the SLAs before the VMs with the less likelihood. We develop this scheduling strategy called SLA-aware scheduling algorithm deployed in "CPack" and adaptive disk scheduling algorithm employed in "DPack." Both algorithms use the probability that each VM will fail its SLA to adjust its priority. The priority is updated at every request completion, allowing "CPack" and "DPack" to quickly adapt to the request arrival fluctuation. "DPack" also dynamically adjusts the resource allocations to reduce the time that a hard disk spends on moving the disk head to the right location before retrieving or storing the data, known as the seek time. Less seek time results in the response time improvement. Traditional hard disks are extensively used to store massive data, e.g. VMs' virtual disk files. To further improve the response time, "CPack" deploys "balance scheduling" algorithm to mitigate the synchronization latency in VMs with more than one virtual CPU (vCPU) or SMP VMs. "Balance scheduling" simply balances the vCPU siblings on different physical CPUs in order to increase the probability that the vCPUs run concurrently. Then, one of the vCPUs that waits for a lock can immediately acquire the lock when the vCPU that holds the lock releases it. This balancing helps reduce the synchronization latency, improving the VM response time. With "Pack[superscript 2]," the VMs run more efficiently, enhancing the response time. The system also quickly manages the CPU and storage resources to achieve the fine-grained SLAs. The results show that "Pack[superscript2]" can achieve 11-69% better average response time than the existing schedulers in KVM when using SMP VMs. "Pack[superscript 2]" is also able to satisfy to all SLAs in the experiments where many VMs are consolidated into the system. But the default KVM CPU and disk schedulers with an appropriate priority tuning can satisfy only two out of four SLAs. When we change from the default disk scheduler to the other disk schedulers available in KVM, all SLAs are violated. In practice, an additional VM would be consolidated into the system if the performance of all VMs still meets the requirements in the SLAs. Without compromising the SLAs, "Pack[superscript 2]" can improve the consolidation ratio by 25%, compared to the default CPU and disk schedulers in KVM. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://bibliotheek.ehb.be:2222/en-US/products/dissertations/individuals.shtml.]