Modeling and Analysis of Virtualized Multi-Service Cloud Data Centers with Automatic Server Consolidation and Prescribed Service Level Agreements

Cloud Data Centers (CDC) are developing rapidly and will have a major impact on IT infrastructures in the future for reasons of their low ramp-up costs and service delivery/support capabilities for the users. In this paper CDCs with multi-service application classes are considered which are operated under an automatic server consolidation based on parallel hysteresis methods for server activations/deactivations which have been reported on our previous work. Each class is subjected to an individual SLA, e.g., for the average service delay for non-real-time services or for delay percentiles for services with strict response time constraints, and probabilities for service rejection (loss) or migration. The CDC is modeled by a multi-class server cluster (SC) system, each of them represented by a multi-server queuing system which is controlled by a Finite State machine (FSM) for each class of cloud services. The SC systems are analyzed exactly under Markovian assumptions to receive averages and percentiles of response times and probabilities of loss or migration. The method is novel as it minimizes the energy consumption for servers by an automatic server consolidation strategy while guaranteeing the negotiated SLAs. The method is based on a worst case boundary consideration for the delays of arriving service requests and can be useful to understand the parametric influences and to assess the energy saving gains for multi-tier CDCs.