Managing Imprecise Worst Case Execution Times on DVFS Platforms

Abstract

Energy efficient real-time task scheduling attracted a lot of attention in the past decade. Most of the time, deterministic execution lengths for tasks were considered, but this model fits less and less with the reality, especially with the increasing number of multimedia applications. It's why a lot of research is starting to consider stochastic models, where execution times are only known stochastically. However, authors consider that they have a pretty much precise knowledge about the properties of the system, especially regarding to the worst case execution time (or worst case execution cycles, WCEC).In this work, we try to relax this hypothesis, and assume that the WCEC can vary. We propose miscellaneous methods to react to such a situation, and give many simulation results attesting that with a small effort, we can provide very good results, allowing to keep a low deadline miss rate as well as an energy consumption similar to clairvoyant algorithms.

EClass: An execution classification approach to improving the energy-efficiency of software via machine learning

“…We modified the source code of the simulator so that it would pause at the end of each interval, output the running values of the first two statistics, and await new frequency settings (if any). In the training phase, f* i,j is computed by solving equation (5) with m = 3 using the roots function in MATLAB. Note that the authors of [7][29] [38][39] also assumed cubic power models.…”

“…EClass then divides S into subsets using a partition clustering algorithm, and the optimal frequency for each interval, denoted as f* i,j , is computed from each value of s i,j based on equation (5). The purpose of discretization is to identify major changes in utilization.…”

“…As discussed in the previous subsection, the statistics sim_diss_* represents P off + P on in equation 4, but in order to solve equation (5) in the training phase, the task-specific parameters P off and C must be determined. Because there are no performance monitors in our test bed that can map individually to these parameters, we need to estimate their values for all frequencies using the energy consumption statistics and execution times collected from training runs.…”

“…Such a usage trend may increase the resource utilization of a running application component. However, the delays incurred when communicating with other components or issuing commands to the underlying operating system followed by receiving the responses may not require a processor to run at a full speed, which provides various opportunities for an energy-aware task scheduling technique [5] [11] [22] to modify the resources allocated to the application component for completing a program execution.…”

Applying Effective Dynamic Frequency Scaling Method in Contactless Smart Card

“…Though DFS method thinking is created from DVFS and it also needs to detect the present module power load, it doesn't need the future power load prediction algorithm as DVFS. DVFS method in [15] combines the long term metric value prediction and duration prediction methods; [16] improves the robustness of low power scheduling algorithms on DVFS by assuming that the worst case execution number of cycles might be imprecisely known; [17] uses a prediction equation based on the correlation of the memory access rate. For the reason of that DFS method does not need to actively change the voltage and does not need to realize the load prediction algorithm as DVFS method before adjusting the clock frequency, its clock switch speed is much quicker than DVFS method.…”