高级检索
    肖 鹏, 胡志刚. 一种面向网格计算的自适应动态冗余预留策略[J]. 计算机研究与发展, 2013, 50(3): 481-489.
    引用本文: 肖 鹏, 胡志刚. 一种面向网格计算的自适应动态冗余预留策略[J]. 计算机研究与发展, 2013, 50(3): 481-489.
    Xiao Peng, Hu Zhigang. An Adaptive Dynamic Redundant Reservation Strategy in Grid Computing[J]. Journal of Computer Research and Development, 2013, 50(3): 481-489.
    Citation: Xiao Peng, Hu Zhigang. An Adaptive Dynamic Redundant Reservation Strategy in Grid Computing[J]. Journal of Computer Research and Development, 2013, 50(3): 481-489.

    一种面向网格计算的自适应动态冗余预留策略

    An Adaptive Dynamic Redundant Reservation Strategy in Grid Computing

    • 摘要: 针对网格环境中冗余机制在系统可靠性和任务执行效率之间难以平衡的问题,提出一种冗余度可动态调整的自适应冗余策略.该策略以可靠性指标为约束条件,依据负载变化自适应地优化系统冗余度,在不降低可靠性的前提下减少冗余度过高对网格任务执行效率的负面影响.理论分析给出了冗余度与可靠性之间的量化关系,实验分析对比了该动态冗余策略与其他冗余策略的性能差异,并对策略关键参数进行了对比分析.实验结果显示,当系统面临较高的任务负载或负载变化剧烈时,自适应的动态冗余预留策略能够显著提高资源有效利用率,并在任务执行效率和系统可靠性之间实现动态平衡,从而降低传统冗余策略对系统性能的负面影响.

       

      Abstract: Redundant mechanism has been widely applied as an effective means to enhance the reliability of grid systems. However, it will lead to higher ineffective utilization, which in turn brings about many negative effects on applications’ performance. To address this problem, an adaptive redundant reservation strategy is proposed with the aim to mitigate the negative effects of conventional strategies. It provides a mechanism that enables the grid systems adjust redundant degree at runtime without decreasing the reliability. The quantitative relationship between redundant degree and reliability is presented theoretically, and extensive experiments based on real-world workload are conducted to examine the performance of the proposed strategy compared with other redundant strategies. The experimental results show that the proposed redundant strategy outperforms other existing strategies in terms of effective resource utilization. Also it brings about tradeoff between the reliability and the execution performance of applications, which significantly mitigates the negative effectives of conventional redundant strategies.

       

    /

    返回文章
    返回