📄 93.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>Codes that could benefit from dynamic load balancing</title>	<meta name="Generator" content="ATutor">	<meta name="Keywords" content=""></head><body> <p>Dynamic load balancing is often difficult to implement with pure MPI, due to the need to rearrange data structures throughout the application or dynamically create/destroy processes. The MPI-2 standard allows for dynamic process manipulation, but this is not yet commonly available. </p>

<p>A multilevel application can spawn a varying number of OpenMP threads from each MPI process as needed to achieve dynamic load balancing. You can control the number of threads assigned to each MPI process through calls to OMP_SET_NUM_THREADS. </p>

<p>As an example, <em>Adaptive Mesh Refinement</em> (AMR) is a technique in which grid resolutions are dynamically increased in regions where physical quantities are rapidly varying. A multilevel AMR code could be structured so that </p>


<ul>
  <li>
    <p>Each MPI process is assigned a segment of the physical domain. Each process would be allowed to increase (or decrease) the grid resolution based on the occurrence of rapidly varying quantities; a variable number of threads 
would then be spawned from each process based on the total volume of computation to achieve approximate load balancing. </p>
  </li>

  <li><p>Exchange of data at the boundaries of the sub-domains would be accomplished via message passing. </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 + -