📄 96.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>Ease of implementation</title> <meta name="Generator" content="ATutor"> <meta name="Keywords" content=""></head><body> <p>In some cases, a multi-dimensional parallel decomposition may result in a more favorable ratio of computation to communication relative to a lower-dimensional decomposition. For example, for a partial differential equation solver in D
space dimensions, slicing up the domain in D different directions results in the highest ratio of calculation to communication per processor. In this situation, a one-dimensional decomposition would not scale to as many processors as would a higher-dimensional decomposition, because the communication becomes dominant
sooner. </p>
<p>However, a multi-dimensional decomposition may be complicated to implement using MPI alone. In such cases a multilevel approach in which a simpler one-dimensional
decomposition is used to construct an MPI application, which is then extended by having MPI processes spawn teams of OpenMP threads, could lead to scaling to higher numbers of processors without the development effort required by the
multi-dimensional implementation. The resulting MLP application would presumably not scale quite as well as the MPI code with a higher-dimensional parallel decomposition,
but the development cost of adding the OpenMP layer to the lower-dimensional code is essentially nil. </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 + -