【摘要】 提出与描述了一种网格环境下任务的执行时间预测的新方法,该方法不需要参考历史数据,可以让用户在提交网格作业之前进行任务执行时间的精确预测。该预测模块使用了R脚本作为基本工具,结合了静态分析、BenchMark数据库解析和基于编译器的方法等3种技术,其中R软件以命令行的方式很容易编写各种类型的任务代码,使其具有自适应性、灵活性的特点,最后分析了预测模块的软件体系结构与组成。功能测试与性能测试结果表明,该预测方法具有正确性、高效性、高精确度,可以给网格的任务调度提供参考,最终提高网格中资源的利用率。更多还原

【Abstract】 A new job execution time prediction module is presented and described,which is used in dynamic and heterogeneous grid environment for promoting the grid resource scheduling efficiency.As there is no historical data available for the prediction module to rely upon when making its predictions,the approach had to rely upon analyzing the behavior of incoming jobs(R script).The execution time of R scripts is estimated by using a combination of static analysis,analytical benchmarking and compiler-based approach.The system architecture and techniques are robust enough such that it allows adaptability and flexibility when applied to other program written using command line paradigm.The software architecture and organization of the prediction module are also discussed in detail subsequently.The experimental results show that this technique is successful in achieving a prediction accuracy of 80% under a normal distribution.更多还原