On the connections between PCTL and dynamic programming
Proceedings of the 13th ACM international conference on Hybrid systems: computation and control
Hi-index | 0.03 |
In this paper, we used the accepted methodology of applying probabilistic-model checking to dynamic power management. For the first time, we analyze in this context a dynamic-power-management system with the two request priorities. First system we consider has a single two-priority service request queue. Second system to be analyzed has two single service request queues that are dedicated for high- and low-priority requests. We calculate optimal stochastic policies for the both systems. The same delay constraint is applied on the whole queue of the first system and on the high-priority (QH) queue of the second system. The comparison of the power consumptions of the two systems shows that the second system consumes up to 77% less power. We obtain that the power consumptions of both systems converge when the constraints become tighter. We demonstrate that, with the increased length of the QH service request queue of the second system, more power can be saved in applications with varying quality-of-service requirements. Our models and experiments can be easily generalized to N request priorities and different queue lengths.