Performance of self-adapting power-saving algorithms for ICT systems

As one of the large energy consumers worldwide, power consumption of network resources can be reduced significantly by dynamic activation/deactivation of resources depending on the current load or state of the system. In this paper, self-adapting algorithms are considered for such resources with/without activation overhead. ICT systems are modeled by a single/multi-server queuing system with finite buffer capacity whose control is performed by a Finite State Machine (FSM) with various hysteresis operation modes. Current system load is represented by the actual state of the ICT system resources, but the method is also applicable for actually measured or forecast load. Queuing systems are analyzed exactly under Markovian assumptions for which a new and efficient recursive algorithm is proposed yielding the probabilities of state and performance values as the mean delay, server activation/deactivation rates, as well as upper bound for the delay distributions. The proposed method allows for optimizing system parameters with respect to given thresholds for the mean delay and to delay percentiles.