面向独立网格任务的松弛预留策略

提出一种支持时间和空间二维松弛的预留接纳策略,允许在一定条件下接纳与已有预留存在叠交的预留请求,从而降低预留服务对系统性能的负面影响。理论分析给出了松弛预留策略的系统收益与风险计算方法,并针对独立任务的协同预留问题设计了相应的请求接纳算法。实验结果表明,松弛预留策略能有效提高资源利用率并降低预留请求拒绝率,当系统面对较高的预留请求率时,松弛预留策略对动态负载变化具有良好的自适应性。

Relaxed reservation policy for parameter sweep applications in computational grid

To mitigate the negative effects brought in by advance reservation, a reservation admission model based on two dimensional relaxed strategy was proposed. The model allowed the acceptance of new reservation requests that overlap with existing ones under certain conditions in grid environment. Both the system benefit and risk of the proposed admission model were theoretically analyzed. Experiment results show that the relaxed reservation strategy can achieve higher resource utilization and lower rejection rate compared with conventional reservation policy and backfilling-based reservation mechanism. In addition, it also shows better self-adaptability to system dynamic reservation change.