The Research Of Low-power Strategies On Real-time Multi-core Systems

Mobile and embedded devices have been indispensable to people’s life. Withthe development of technology, the performance of these devices is getting higherand higher, along with which comes the problem of too short battery life. Limited bythe materials, the update rate of the battery is much slower than that of the devices.And the development of new battery needs a very high input. Previous researchesdemonstrate that low-power techniques can effectively reduce energy consumptionof the systems and make devices satisfy users better.DPM and DVS are the most extensively studied low-power techniques. DPMtechnique is based on the modularization of the systems, which can turn the moduleson or off automatically so as to reduce power consumption. DVS technique aims atcontrolling the power consumption of the systems by adjusting the executing speeddynamically. This paper focus on the research of DVS technique on real-timemulti-core systems, meanwhile DPM technique is also taken into consideration.We have proposed two DVS strategies based on EDF scheduling，which areCC-SE-DVS and CC-R-E-DVS. CC-SE-DVS strategy is based on CC-EDF-DVS, andwe have put forward a concept of segment in this strategy. A segment is defined as aperiod of time which starts at the end of an idle time and ends at the beginning of thenext idle time of the processor. Voltage adjusting in current segment would not beaffected by the previous one, thus the needless effect from the previous segment canbe eliminated and the power consumption can be efficiently reduced. CC-R-E-DVSis based on CC-SE-DVS and it can make use of the slack time from previous task. Ithas the characteristic of CC-EDF-DVS and CC-RM-DVS, and it can further reducethe power consumption of the system. The simulation results show that the strategieswe have proposed are more efficient than previous ones, meanwhile, they areadaptive for all kinds of task distribution.We have constructed a multi-core platform on SIM-WATTCH, and we also havestudied task scheduling in multi-core systems. We have put forward a multi-core taskscheduling algorithm which can assign the tasks to the optimal core. The schedulingalgorithm assigns the tasks properly by matching the time utilization, so it caneffectively adjust the time utilization of each processor core and turn off the idlecore timely. The simulation results show that the multi-core scheduling algorithm we have proposed can balance the task load of the system better, so it can createconditions for voltage scaling and further reduce energy consumption of the system.