AUTONOMIC MANAGEMENT OF SOFTWARE AS A SERVICE SYSTEMS WITH MULTIPLE QUALITY OF SERVICE CLASSES

In recent years the emergence of Software as a Service (SaaS) provision and cloud computing in general had a tremendous impact on corporate information technology. While the implementation and successful operation of powerful information systems continues to be a cornerstone of success in modern enterprises, the ability to acquire IT infrastructure, software, or platforms on a pay-as-you-go basis has opened a new avenue for optimizing operational costs and processes. In this context we target elastic SaaS systems with on-demand cloud resource provisioning and implement an autonomic management artifact. Our framework forecasts future user behavior based on historic data, analyzes the impact of different workload levels on system performance based on a non-linear performance model, analyzes the economic impact of different provisioning strategies, derives an optimal operation strategy, and automatically assigns requests from users belonging to different Quality of Service (QoS) classes to the appropriate server instances. More generally, our artifact optimizes IT system operation based on a holistic evaluation of key aspects of service operation (e.g., system usage patterns, system performance, Service Level Agreements). The evaluation of our prototype, based on a real production system workload trace, indicates a cost-of-operation reduction by up to 60 percent without compromising QoS requirements