首页> 外文会议>2nd international conference on bioinformatics and computational biology 2010 >Study of Computing Consolidation Techniques in Computational Protein Loop Structure Modeling
【24h】

Study of Computing Consolidation Techniques in Computational Protein Loop Structure Modeling

机译:计算蛋白质环结构建模中的计算整合技术研究

获取原文
获取原文并翻译 | 示例

摘要

In this article, we advocate the approach of "computing consolidation" to achieve efficient usage of computer resources in protein structural modeling applications. The fundamental idea of computing consolidation is to increase computational density to a processor in order to increase CPU and other resources utilization rate. We use our computational protein loop structure modeling application with clear memory-intensive and computation-intensive energy function evaluation components as an example to investigate the effectiveness of computing consolidation. We study various implementations of computing consolidation, including process consolidation, fine-grain and coarse-grain thread consolidations on a variety of hardware architectures. Our computational results show that thread consolidation leads to more significant speedup due to the cache effect and the lower cost in thread management than process management. Moreover, different thread consolidation implementations are suitable for different hardware architectures. By consolidating threads carrying out energy function components with heterogeneous resource demands, energy-function-level thread consolidation outperforms iteration-level thread consolidation with over 10% speed improvement in single-CPU processors. However, on hyper-threading-enabled processors or multi-core processors, iteration-level thread consolidation enables more computation overlaps in the concurrent threads carrying out independent iterations, which yields more aggressive computation speed improvement.
机译:在本文中,我们提倡“计算合并”的方法来实现蛋白质结构建模应用程序中计算机资源的有效利用。计算合并的基本思想是增加处理器的计算密度,以提高CPU和其他资源的利用率。我们将计算蛋白质环路结构建模应用程序与清晰的内存密集型和计算密集型能量函数评估组件作为示例,以研究计算整合的有效性。我们研究了各种计算合并的实现,包括各种硬件体系结构上的流程合并,细粒度和粗粒度线程合并。我们的计算结果表明,与进程管理相比,由于缓存效果和线程管理成本较低,线程合并导致更显着的加速。此外,不同的线程合并实现适用于不同的硬件体系结构。通过合并执行具有异构资源需求的能源功能组件的线程,能源功能级线程合并的性能优于迭代级线程合并,并且单CPU处理器的速度提高了10%以上。但是,在启用了超线程的处理器或多核处理器上,迭代级线程合并使执行独立迭代的并发线程中的计算重叠更多,从而大大提高了计算速度。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号