📄 92.html

📁 国外MPI教材
💻 HTML
字号:
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"    "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" lang="en"><head>	<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1" />	<style type="text/css">	body { font-family: Verdana, Arial, Helvetica, sans-serif;}	a.at-term {	font-style: italic; }	</style>	<title>Example: NAS Parallel Benchmarks</title>	<meta name="Generator" content="ATutor">	<meta name="Keywords" content=""></head><body> <ul>
</ul>
<p>This example is of a standard set of benchmarks for massively parallel systems. They simulate the computation and communication patterns of typical large-scale applications in computational fluid dynamics. Overall performance of the various benchmarks are influenced strongly by the detailed balance between the principal 
hardware components: </p>

<ul>
  <li> 
    <p>CPU</p>
  </li>
  <li> 
    <p>memory (and cache)</p>
  </li>
  <li> interconnect</li>
</ul>
<p>A detailed comparison between a multilevel implementation and pure MPI shows that the relative performance depends in a complicated way on the: </p>

<ul>
  <li>
    <p>application</p>
  </li>
  <li>
    <p>size of the dataset</p>
  </li>
  <li>
    <p>underlying hardware configuration</p>
  </li>
</ul>

<p>In many cases the pure MPI implementation performs best; however, many instructive examples for which the multilevel performance is superior can be found among the various cases. </p>

<ul>
  <ul>
  </ul>
  <li>

    <p>In one case (the CG benchmark), the pure MPI code involves relatively large messages, the number of which grows strongly with the number of processors. At a critical point (around 8 processors for small datasets; even lower 
for larger datasets), the communication overhead becomes so significant that the multilevel implementation, with fewer MPI processes for a given total number of processors, performs better. </p>
</li>

  <li><p>In the FT benchmark, "all-to-all" collective communications rapidly come to dominate as the processor count is increased. Again, the multilevel implementation 
allows one to go to higher processor counts while restricting the number of MPI processes, leading to improved scaling in this regime. </p>
  </li>
</ul></body></html>
⌨️ 快捷键说明

复制代码 Ctrl + C
搜索代码 Ctrl + F
全屏模式 F11
切换主题 Ctrl + Shift + D
显示快捷键 ?
增大字号 Ctrl + =
减小字号 Ctrl + -