Quality, Reliability, Security and Robustness in Heterogeneous Systems. 13th International Conference, QShine 2017, Dalian, China, December 16 -17, 2017, Proceedings

Research Article

Classification-Based Reputation Mechanism for Master-Worker Computing System

Download
127 downloads
  • @INPROCEEDINGS{10.1007/978-3-319-78078-8_24,
        author={Kun Lu and Jingchao Yang and Haoran Gong and Mingchu Li},
        title={Classification-Based Reputation Mechanism for Master-Worker Computing System},
        proceedings={Quality, Reliability, Security and Robustness in Heterogeneous Systems. 13th International Conference, QShine 2017, Dalian, China, December 16 -17, 2017, Proceedings},
        proceedings_a={QSHINE},
        year={2018},
        month={4},
        keywords={Node classification Reinforcement learning Reputation system},
        doi={10.1007/978-3-319-78078-8_24}
    }
    
  • Kun Lu
    Jingchao Yang
    Haoran Gong
    Mingchu Li
    Year: 2018
    Classification-Based Reputation Mechanism for Master-Worker Computing System
    QSHINE
    Springer
    DOI: 10.1007/978-3-319-78078-8_24
Kun Lu1,*, Jingchao Yang1, Haoran Gong1, Mingchu Li1
  • 1: Dalian University of Technology
*Contact email: lukun@dlut.edu.cn

Abstract

Master-worker computing is a parallel computing scheme, which makes master and worker collaborate. Due to its high reliability availability and serviceability, it is widely used in scientific computing fields. However, lack of cooperation and malicious attack in Master-worker computing can greatly reduce the efficiency of parallel computing. In this paper, we consider a reputation system based on individual classification to inducing worker nodes returning true answer and separate malicious worker nodes. By introducing reinforcement learning, rational workers are induced to behave cooperatively and auditing rate of the master decreases. Our model is based on evolutionary game theory. Simulation results show that our reputation system can not only effectively guarantee eventual correctness, separate malicious worker nodes, but also save the master node’s auditing cost.