Scholarlink

<p>Dynamic Voltage and Frequency Scaling (DVFS) platforms are highly effective in reducing energy consumption, as they allow the adjustment of operating frequency and voltage in predefined pairs. Lowering a task’s execution frequency proportionally reduces the voltage, thereby decreasing overall energy usage. This article presents the design of a low-complexity, energy-efficient frequency scaling mechanism for executing dependent tasks on DVFS-enabled platforms. In such applications, energy can be minimized using the Scaling Axiomatic Approach (<i>SAA</i>), which reduces task frequencies while respecting dependency constraints. However, this method suffers from high time complexity, often exceeding that of standard scheduling algorithms, which limits its practicality. To address this, we introduce <i>GinGa</i>, a lower-bound frequency scaling mechanism for DVFS-enabled platforms that serve as an efficient replacement for <i>SAA</i>. <i>GinGa</i> achieves energy savings close to those of <i>SAA</i>, with only minor performance degradation. <i>GinGa</i> was evaluated on a large set of randomly generated task graphs and three real-world applications, demonstrating its ability to effectively replace <i>SAA</i> while retaining most of its energy-saving benefits. Additionally, <i>GinGa</i> can be combined with slack extender mechanisms to further enhance frequency scaling. These results highlight <i>GinGa</i> as a low-complexity, efficient solution for energy-aware scheduling in DVFS-enabled platforms.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A replacement for the scaling axiomatic approach to scheduling dependent tasks on DVFS computing platforms

  • 期刊论文

  • 发表日期：

    2025年7月21日

查看全文
  • Tarek Hagras,
  • Gamal A. El-Sayed

摘要

Dynamic Voltage and Frequency Scaling (DVFS) platforms are highly effective in reducing energy consumption, as they allow the adjustment of operating frequency and voltage in predefined pairs. Lowering a task’s execution frequency proportionally reduces the voltage, thereby decreasing overall energy usage. This article presents the design of a low-complexity, energy-efficient frequency scaling mechanism for executing dependent tasks on DVFS-enabled platforms. In such applications, energy can be minimized using the Scaling Axiomatic Approach (SAA), which reduces task frequencies while respecting dependency constraints. However, this method suffers from high time complexity, often exceeding that of standard scheduling algorithms, which limits its practicality. To address this, we introduce GinGa, a lower-bound frequency scaling mechanism for DVFS-enabled platforms that serve as an efficient replacement for SAA. GinGa achieves energy savings close to those of SAA, with only minor performance degradation. GinGa was evaluated on a large set of randomly generated task graphs and three real-world applications, demonstrating its ability to effectively replace SAA while retaining most of its energy-saving benefits. Additionally, GinGa can be combined with slack extender mechanisms to further enhance frequency scaling. These results highlight GinGa as a low-complexity, efficient solution for energy-aware scheduling in DVFS-enabled platforms.