arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1902.10069cs.DC
收藏

数据网格中计算作业的数据访问模式模拟

Simulating Data Access Profiles of Computational Jobs in Data Grids

Volodimir Begy, Joeri Hermans, Martin Barisits, Mario Lassnig, Erich Schikuta

摘要:

由于近年来高速局域网和广域网的迅速发展,在计算网格中运行的应用程序的数据访问模式正在发生变化。不再严格要求数据密集型作业在相应输入数据所在的计算站点处运行。相反,作业可以使用数据放置、分段和远程数据访问的任意组合来访问数据。这些数据访问配置文件表现出部分不重叠的吞吐量瓶颈。可以利用这一事实来最小化作业等待输入数据的时间。在这项工作中,我们提出了一种新的网格计算模拟器,它把重点放在各种数据访问配置文件。基于我们的模拟器的基本假设是合理的,在欧洲核子研究中心的全球大型强子对撞机计算网格(WLCG)进行的经验实验。我们演示了如何校准模拟器的参数,根据真实的系统使用后验推理与似然无马尔可夫链蒙特卡罗。此后,我们验证了模拟器的输出相对于一个真实的生产工作量从WLCG,证明其显着的准确性。

英文摘要:

The data access patterns of applications running in computing grids are changing due to the recent proliferation of high speed local and wide area networks. The data-intensive jobs are no longer strictly required to run at the computing sites, where the respective input data are located. Instead, jobs may access the data employing arbitrary combinations of data-placement, stage-in and remote data access. These data access profiles exhibit partially non-overlapping throughput bottlenecks. This fact can be exploited in order to minimize the time jobs spend waiting for input data. In this work we present a novel grid computing simulator, which puts a heavy emphasis on the various data access profiles. The fundamental assumptions underlying our simulator are justified by empirical experiments performed in the Worldwide LHC Computing Grid (WLCG) at CERN. We demonstrate how to calibrate the simulator parameters in accordance with the true system using posterior inference with likelihood-free Markov Chain Monte Carlo. Thereafter, we validate the simulator's output with respect to an authentic production workload from WLCG, demonstrating its remarkable accuracy.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑