Are task mappings with the highest frequency of servers so good? A case study on Heterogeneous Earliest Finish Time (HEFT) algorithm

Abstract

In the current heterogeneous computing environment, scheduling strategies play a key role in achieving high performance under current heterogeneous computing environment. Various algorithms have been proposed to generate high-quality schedules. Nonetheless, all the existing methods initialize task mapping with the highest frequencies of servers, creating significant effects on the results. Does mapping with the highest frequencies of servers generate the best schedule? In this paper, an attempt has been made to answer this question by investigating HEFT, which is a commonly used initial schedule method in scheduling the literature under DVFS environments. In this work, the task mapping is initialized with different server frequency level, along with investigating its impacts on the schedule results. Based on this concept, a corresponding Frequency Scaling Algorithm (FSA) has been proposed. By applying FSA combined with three frequency levels mappings, it has been observed that initial task mapping with middle server frequencies can generate much better results. When the initial frequency is the highest, the scheduling result is not optimal. The proposed algorithm requires less execution time and generates stable performance. Also, the proposed frequency scaling method is found to mitigate the impacts of initial mappings to a certain extent.