A study on periodic shutdown for adaptive CMPs in handheld devices

The challenge to satisfy the demand for higher computing performance has become an increasingly difficult task to achieve. In the area of mobile devices, this demand has to be carefully balanced with an efficient use of the power source. We propose the use of an adaptive architecture that enables savings in power and energy in an intuitive way, considering the properties of future process technologies. We satisfy performance demand by utilizing thread-level parallelism and minimize the power and energy consumption by proposing an adaptive strategy that manages the power state of each individual CMP-core. In this study, we propose a periodical shutdown strategy and evaluate it in a multiprogrammed workload environment. Results show that a large amount of idle time, 77 %, can be saved by putting processors into power-saving states. Furthermore, introducing timeouts can dramatically decrease the number of state transitions.