📄 99.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>Scaling Performance: Intel Cluster</title>	<meta name="Generator" content="ATutor">	<meta name="Keywords" content=""></head><body> <p>On the OSC Intel cluster, the scaling behavior of the pure MPI implementation shows clearly the limiting effects of intra-node memory bandwidth: <br/>
  <br/>
  <img src="Cluster-MPI.gif" align=center> </p>

<p>The scaling is noticeably and uniformly worse when multiple PEs within a node are used. Indeed, with four PEs per node the performance with 96 processors is <em>worse</em> than can be achieved with 24 processors in separate nodes. </p>

<p>The following plots compare the pure MPI code to the MLP implementation as before. In each case we see that on this system the MPI implementation tends to be better for moderate numbers of processors, but eventually there is a crossover where the MLP implementation becomes superior. Hence MLP again allows us to effectively scale the code to more processors than does pure MPI and so obtain 
  a faster solution of the problem. <br/>
  <br/>
  <img src="Cluster-2PEs.gif" align=center> <br/>
  <br/>
  <img src="Cluster-4PEs.gif" align=center> </p></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 + -