A stochastic DVS-based dynamic power management for soft real-time systems

This paper introduces a stochastic dynamic power management policy for soft real-time systems which uses DVS to consume less power in the main processor while satisfying the performance constraints. The idea is based on a Markovian model of the system. The system under study is a single processor system with real-time jobs which arrive according to a Poisson process and have exponentially distributed deadlines and service times. Since the speed levels in real processors are not continuous, the required service rate for a given performance is estimated in a probabilistic manner using its two immediate neighboring speeds. Experimental results show that the long run power consumption of the system improves even comparing to a fixed-speed system which uses a continuous model of processor speeds, while all of the real-time constraints are satisfied.