Automatic energy efficiency management of data center resources by load-dependent server activation and sleep modes

The operation of large Data Centers (DC) with thousands of servers is very costly in terms of energy consumption and cooling requirements. Currently, major efforts can be observed for server virtualization and consolidation to approach a proportionality between computation amount and energy consumption. In this contribution, a generalized model is presented which allows an automatic server consolidation by a load-dependent control of server activations using multi-parallel hysteresis thresholds, cold and hot server stand-by, and Dynamic Voltage and Frequency Scaling (DVFS). For the energy-efficiency and performance analysis, a multi-server queuing model is defined which is controlled by a Finite State Machine (FSM). The parameters of the queuing model are defined such that Service Level Agreements (SLA, e.g. as mean or percentiles of response times) are guaranteed except for overload conditions. The queuing model can be exactly analyzed under Markovian process assumptions from which all relevant quality of service (QoS) and energy efficiency (EE) metrics are derived. Numerical results are provided which demonstrate the applicability of the proposed model for the DC management, in particular to theoretically quantify the tradeoff between the conflicting aims of EE and QoS.