Reliability-aware task scheduling for energy efficiency on heterogeneous multiprocessor systems

Recent studies mainly focus on high performance or low power consumption for task scheduling on heterogeneous multiprocessor systems (HMSs). Dynamic voltage and frequency scaling (DVFS) is an important energy reduction technique, which adjusts the voltage and frequency of the processor while the tas...

Full description

Saved in:
Bibliographic Details
Published in:The Journal of supercomputing 2021-10, Vol.77 (10), p.11643-11681
Main Authors: Deng, Zexi, Cao, Dunqian, Shen, Hong, Yan, Zihan, Huang, Huimin
Format: Article
Language:English
Subjects:
Algorithms
Compilers
Computer Science
Constraints
Critical path
Electric potential
Failure rates
Interpreters
Machine learning
Microprocessors
Multiprocessing
Optimization
Performance enhancement
Power consumption
Processor Architectures
Programming Languages
Scheduling
System reliability
Task scheduling
Voltage
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Recent studies mainly focus on high performance or low power consumption for task scheduling on heterogeneous multiprocessor systems (HMSs). Dynamic voltage and frequency scaling (DVFS) is an important energy reduction technique, which adjusts the voltage and frequency of the processor while the task is executing. However, some studies have shown that reducing the voltage of processor increases the transient failure rate, which reduces system reliability. In this paper, we aim at addressing the scheduling problem of optimizing energy under makespan and reliability constraints on HMSs with DVFS. We first propose an improved whale optimization algorithm (WOA) deploying opposition-based learning and individual selection strategy, which can balance the exploration and exploitation ability. To maintain population diversity, we then apply a constrained rank-based method which retains some infeasible individuals in the population. Finally, we reschedule the Critical Path Nodes (CPNs) to further improve the performance of improved WOA. The main difference between our work and most previous works is that we study a new scheduling problem, and utilize an improved WOA algorithm integrating with rescheduling CPNs and a constrained rank-based method. Extensive experiments are conducted to evaluate our proposed algorithm, and the evaluation results show that our proposed algorithm is compelling in comparison with the state-of-the-art algorithms.
ISSN:0920-8542
1573-0484
DOI:10.1007/s11227-021-03764-x