1. In the proof of Theorem 1, no on/off switching energy overhead is considered. However, in the Intel Xscale processor model used in the experiment part, there is switing energy overhead for powering down. When this energy overhead is considered, theorem 1 might not be true. This needs to be justified in the proof of theorem 1.
Despiteits simplicity, this power model captures the essential components for the system-wide energymanagement. Here, is the static power, which includes the power to maintain basic circuits and keep the clock running and can be removed only by powering off the whole system. Due to the prohibitive overhead of turning off/on a system in periodic real-time execution settings, we assume that the system is in on state at all times and that always consumes and is not manageable. Different will not affect the absolute energy savings and, for simplicity, we will ignore the static power(i.e, ).
2. The authors declared in the response to Reviewer 2 that the algorithm DSRDP is based on the DAC'05 paper. But the author's explanation on DSRDP on Section 4 does not reflect that completely: First, DAC'05 paer is not cited here and the author did not mention in the experiment section that DSRDP is based on DAC'05 paper, either. Second, even assuming DSRDP is based on DAC'05 paper, the authors' explanation on it in Section 4 is not consistent with the algorithm in DAC'05 paper. Note that the algorithm in DAC'05 is not simply based on improvement of DRA algorithm in [1], they are much different in terms of both slack time calculation and the way of reclaiming it. Moreover, it try to procrastinate the tasks not just when the processor is in idle status . They also use dynamic procrastination which can procrastinate the task even when the processor is not idle. The authors need to revise DARDP based on DAC'05 paper and show the actual improvement based on it.