📄 82.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>Hardware and Application Considerations</title>	<meta name="Generator" content="ATutor">	<meta name="Keywords" content=""></head><body> <p>MLP is a very natural approach for writing applications for distributed shared-memory architectures such as clusters of SMPs, Beowulfs, etc. In particular, OpenMP (or other similar approaches) has the potential to be more efficient for certain problems than MPI in the shared memory domain, i.e., within SMP nodes. However, it is difficult to make general statements about the performance of MLP codes 
versus pure MPI or OpenMP because of the wide variety in hardware and application characteristics. </p>

<p>The most important hardware characteristics are </p>

<ul>
  <li>CPU speed</li>
  <li>Memory bandwidth within the SMP node</li>
  <li>Cache characteristics and performance</li>
  <li>Interconnect performance (latency, bandwidth) </li>
</ul>

<p>Applications can vary widely in terms of </p>

<ul>
  <li>their communication patterns, i.e., number of messages, size of messages, how the message traffic varies as the number of processors is increased, etc. </li>
  <li>the amount of synchronization required</li>
  <li>the relative proportion of coarse- to fine-grained parallelism </li>
</ul>

<p>All of these factors can play a role when trying to decide if an MLP approach will be beneficial. In addition, considerations other than pure performance may in some cases favor an MLP approach. </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 + -