📄 95.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 make use of replicated data</title> <meta name="Generator" content="ATutor"> <meta name="Keywords" content=""></head><body> <ul>
<p><li>In pure MPI implementations, such applications are often limited by the available memory.</p>
<ul>
<p><li>This is effectively the total SMP node memory divided by the number of processors per node.</p>
</ul>
<p><li>They may also involve expensive collective operations.</p>
<p><li>A multilevel approach in which, say, a single MPI process is allocated to each node, and teams of OpenMP threads are spawned within nodes, can allow for larger problems to be studied because data replication then happens on a per-node basis.</p>
<p><li>In addition, as before, restricting the number of MPI processes can reduce the communication overhead arising from collective operations.</p>
</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 + -