Adaptive Quality of Service violation detection and remediation for cloud computing

Cloud computing has gained popularity among service providers and consumers to perform business operations due to cost-effectiveness, on-demand, ease of communication, and transaction convenience in terms of accessibility and availability. However, due to the vulnerability of cloud elastic, multi-tenant, and resource-pooling environment, it is crucial to have a binding agreement among all the service parties to guarantee trust while fulfilling the Quality of Services (QoS) agreed in Service Level Agreements (SLA). Cloud applications consume enormous resources due to variations in service users’ demands and needs, raising the issue of over-provisioning expensive cloud resources. Thus, creating a balanced tradeoff between specified QoS terms stated in the SLA with scalable resource allocation is crucial. This thesis focuses on cloud QoS issues and conducts a literature review to understand the potential problem in existing research on QoS metrics, monitoring, QoS violation detection, and remediation. To solve the problem of QoS violation detection, we propose a model to monitor QoS metrics and performance measures to verify compliance with the respective SLAs. An adaptive fuzzy-based model called "Detection Fuzzy Model for QoS Violation" (DeFQoV) is proposed for QoS violation detection using fuzzy 16 decision if-then rules to determine the quality status. Once a violation is detected, a proposed violation remediation model called "Remediation quantified Scaling for QoS Violation" (ReSQoV) using resource planning and scaling is triggered. This ReSQoV model is proposed based on the Universal Scalability Law (USL), which generates a capacity model for specific workloads and cloud resources. ReSQoV considers the system overheads, i.e., contention and coherence, while allocating resources to maintain the agreed QoS. The model DeFQoV is validated with scenarios on response time and throughput for detecting QoS violations to make accurate decisions for appropriate remedial action. Experiment results show the DeFQoV model’s comparison with other classifiers, with less than 0.5% Root Mean Squared Error (RMSE), thus confirming more than 99% detection accuracy for QoS violation. Consequently, as the QoS violation detection decision is "Probably Violation" and "Definitely Violation," the remedial action ReSQoV is triggered to add required resources to the Virtual Machine (VM) as vertical scaling. Experiment results show the ReSQoV model’s comparison with the policy-based resource allocation through the emulated QoS parameters and respective resource utilization scenarios. The results show that QoS is regained after the ReSQoV quantified resource allocation. The model’s validation is performed through the statistical test Analysis of Variance (ANOVA), which shows the significant difference between the policy-based and quantified scaling systems with a p-value less than 5% among the generated QoS parameter values and their interactions