Revenue maximization problems in commercial data centers

As IT systems are becoming more important everyday, one of the main concerns is that users may face major problems and eventually incur major costs if computing systems do not meet the expected performance requirements: customers expect reliability and performance guarantees, while underperforming systems loose revenues. Even with the adoption of data centers as the hub of IT organizations and provider of business efficiencies the problems are not over because it is extremely difficult for service providers to meet the promised performance guarantees in the face of unpredictable demand. One possible approach is the adoption of Service Level Agreements (SLAs), contracts that specify a level of performance that must be met and compensations in case of failure. In this thesis I will address some of the performance problems arising when IT companies sell the service of running ‘jobs’ subject to Quality of Service (QoS) constraints. In particular, the aim is to improve the efficiency of service provisioning systems by allowing them to adapt to changing demand conditions. First, I will define the problem in terms of an utility function to maximize. Two different models are analyzed, one for single jobs and the other useful to deal with session-based traffic. Then, I will introduce an autonomic model for service provision. The architecture consists of a set of hosted applications that share a certain number of servers. The system collects demand and performance statistics and estimates traffic parameters. These estimates are used by management policies which implement dynamic resource allocation and admission algorithms. Results from a number of experiments show that the performance of these heuristics is close to optimal.