In today's digital age, service- and cloud-based applications have become increasingly dynamic, requiring runtime system adaptations to manage their complex behavior. To meet this need, cloud computing offers an elastic infrastructure that can dynamically adjust resources and scale applications as needed. However, as cloud-based systems become more complex, manual management of these systems becomes increasingly challenging and cost-ineffective. To ensure that cloud resources are dynamically reconfigured to meet Quality of Service (QoS) requirements, various research studies have focused on different approaches, such as architecture-based reliability modeling, empirical reliability or resilience assessment, architecture-based performance prediction, and performance analysis in cloud-based systems. Additionally, self-adaptive systems have been developed that use Monitor, Analyze, Plan, Execute, and Knowledge (MAKE-K) loops and similar approaches to realize adaptations, while autoscalers and cloud elasticity promise to maintain stable QoS measures even when workload intensity changes. However, a higher level of abstraction is necessary to make the reconfiguration process of cloud-based systems automatic. This abstraction models the various available technologies and options, which allows us to capture the domain knowledge needed to make informed decisions when choosing optimal reconfiguration solutions. Additionally, empirical research is crucial for supporting the scientific method and creating trust in new technologies and approaches. In this doctoral thesis, we have designed and performed multiple experiments to model and understand different QoS requirements, including reliability, performance, and system overload. We have modeled these measurements analytically and statistically, then validated our models empirically. Through this empirical research, we have investigated the trade-offs of different QoS metrics and studied how these requirements affect reliability and performance in centralized or distributed systems. After gathering empirical evidence, we have used our models for multi-criteria optimization analyses, which give the system self-management ability. The system can assess the situation and automatically adapt cloud resources, choosing an optimal reconfiguration solution. To support this approach, we have focused on capturing domain knowledge in the context of QoS requirements and abstracting available technologies to study them at an architectural level of abstraction. We have also provided architectural analysis tools for self-adaptive service- and cloud-based dynamic-routing systems. Moreover, we have presented a multifaceted reconfiguration of dynamic routing and studied scenarios where components are idle, steady, and transient, as well as the interplay of these scenarios. This dissertation demonstrates the importance of empirical research in creating trust in new technologies and approaches for managing complex cloud-based dynamic-routing systems. By modeling and understanding different QoS requirements and their trade-offs, we have developed a self-management system that can dynamically adapt cloud resources to ensure stable QoS metrics, even in the face of changing workload intensity.

Top