📄 11.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>Parallel Computer Architectures</title>	<meta name="Generator" content="ATutor">	<meta name="Keywords" content=""></head><body> <p> Historically, there have been two types of parallel computer architectures which were differentiated largely by their memory layouts: <em><a href="../glossary.html#shared+memory" target="body" class="at-term">shared memory</a></em> and <em>distributed memory</em>. </p>

<h3> <a name="smp"> Shared Memory </a> </h3>

  
<p> In a <em><a href="../glossary.html#Shared+Memory+System" target="body" class="at-term">shared memory system</a></em>, all <em><a href="../glossary.html#Processors" target="body" class="at-term">processors</a></em> in the system share equal access to one or more banks of memory as shown in Figure 1.1. All areas in memory are equally accessible to all processors and processor-to-processor data transfers are done using shared areas in memory. </p>

<table border="0" width="602">
  <tr>
    <td><img src="1.1.sharedmemory.jpg" align="left" width="602" height="422"></td>
  </tr>
  <tr>
    <td><p><b>Figure 1.1:</b> Example of a Shared Memory System</p></td>
  </tr>
</table>

<p>Examples of shared memory systems include the Cray Y-MP, the SGI Challenge, the Sun UltraEnterprise 10000, and most dual- and quad-processor workstations and PCs.
</p>

<h3> <a name="mpp"> Distributed Memory </a> </h3>

  
<p> In a <em><a href="../glossary.html#distributed+memory" target="body" class="at-term">distributed memory</a></em> system, each processor in the system has its own separate bank or banks of memory, as shown in Figure 1.2. Processor-to-processor data transfers are done over some form of internal network topology. A given processor's 
memory may or may not be visible to other processors, depending on the system design. </p>

<table width="642" border="0">
  <tr>
    <td><img src="mpp.jpg" width="642" height="462"></td>
  </tr>
  <tr>
    <td><b>Figure 1.2:</b> Example of a Distributed Memory System</td>
  </tr>
</table>


<p>Examples of distributed memory systems include the Cray T3D and T3E, the IBM SP2, and clusters of uniprocessor workstations or PCs. </p>

<h3> <a name="dsm"> Distributed Shared Memory </a> </h3>

<p> However, one can easily imagine a hybrid of these approaches, where the "box" containing a a processor and its local memory in Figure 1.2 is replaced with a shared memory system, as shown in Figure 1.3. This is known as a <em><a href="../glossary.html#distributed+shared+memory" target="body" class="at-term">distributed shared memory</a> (DSM)</em> system. </p>

<table width="722" border="0">
  <tr>
    <td><img src="dsm.jpg" height="482" width="722"></td>
  </tr>
  <tr>
    <td><b>Figure 1.3:</b> Example of a Distributed Shared Memory System</td>
  </tr>
</table>

<a name="numa">
<p>DSM systems typically come in one of two "flavors", depending on how they present the hybrid nature of their memory layout to the user. The first flavor of DSM systems attempts to hide the distributed nature of the distributed shared memory by presenting it as a single shared address space in which some locations in memory are closer to a given processor and thus take less time to access than others.This is known as a <em><a href="../glossary.html#non-uniform+memory+access+%5C%28NUMA%5C%29" target="body" class="at-term">non-uniform memory access (NUMA)</a></em> architecture. Examples of NUMA systems include the HP Exemplar, the SGI Origin 2000 and 3000 series, and the Compaq GS320.
</p> 
</a>

<a name="clusterofsmps">
<p> The second flavor of DSM systems does not try to hide the fact that memory is distributed from the user; instead, it presents the user with a distributed memory system of nodes in which each node happens to be a shared memory system. This is known as a <em><a href="../glossary.html#cluster+of+SMPs" target="body" class="at-term">cluster of SMPs</a></em> architecture. Examples of clusters of SMPs include the IBM SP3, the Compaq AlphaServer SC, and the now ubiqitous Beowulf clusters of dual-processor [or multi-core] x86 systems. </p>
</a></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 + -