Computer clusters bring high performance as well as large energy consumption. Energy-efficient scheduling strategies for parallel applications running on a homogeneous cluster can perform efficiently in conserving energy. In order to achieve the goal of optimizing performance and energy efficiency in clusters, we propose an energy-efficient Dependency-based task Grouping (DG) method to assign parallel tasks under precedence constrains to multi-core processors. Dependency degree is defined as the sum of the reduced communication time by assigning task paths with much intercommunication to one processor and the execution time of unexecuted redundant tasks on the same node. Our algorithms aim at reducing energy consumption and improving resource utilization by assigning the task paths with highest dependency degrees to one processor. Combining three existing schedule algorithms-TDS (Task Duplication Scheduling), EAD (Energy-Aware Duplication) and PEBD (Performance-Energy Balanced Duplication) with the DG method, we propose three improved algorithms-TDS-DG, EAD-DG and PEBD-DG. Compared with the three existing algorithms, the improved algorithms can save energy and improve computing resource utilization by 55.4% and 71.2% on average, respectively, at the cost of a slightly 2% performance degradation.
展开▼