blocks|key|3750186|text|我不确定您是否有多台机器，以及您希望如何分配进程的确切细节，但我会考虑阅读：|type|unstyled|depth|inlineStyleRanges|entityRanges|data|3750187|mpirun+man+page|offset|length|3750188|该手册指出，它有将进程绑定到不同事物的方法，包括节点、套接字和cpu核心。|3750189|重要的是要注意，如果您运行的进程数是CPU核心数的两倍，那么您将实现这一点，因为它们往往平均分布在核心上，以分担负载。|3750190|我会尝试下面这样的东西，尽管手册有点模棱两可，而且我不能百分之百地确定它是否会像预期的那样工作，只要你有一个双核心：|3750191|mpirun+-np+4+-npersocket+4+./application|code-block|syntax|javascript|3750192|entityMap|0|LINK|mutability|MUTABLE|url|http://www.open-mpi.org/doc/v1.4/man1/mpirun.1.php^0|0|0|F|0|0|0|0|0|0^^$0|@$1|2|3|4|5|6|7|Y|8|@]|9|@]|A|$]]|$1|B|3|C|5|6|7|Z|8|@]|9|@$D|10|E|11|1|12]]|A|$]]|$1|F|3|G|5|6|7|13|8|@]|9|@]|A|$]]|$1|H|3|I|5|6|7|14|8|@]|9|@]|A|$]]|$1|J|3|K|5|6|7|15|8|@]|9|@]|A|$]]|$1|L|3|M|5|N|7|16|8|@]|9|@]|A|$O|P]]|$1|Q|3|-4|5|6|7|17|8|@]|9|@]|A|$]]]|R|$S|$5|T|U|V|A|$W|X]]]]

I'm not sure if you have multiple machines or not, and the exact details of how you want the processes distributed, but I'd consider reading up:

<a href="http://www.open-mpi.org/doc/v1.4/man1/mpirun.1.php" rel="nofollow">mpirun man page</a>

The manual indicates that it has ways of binding processes to different things, including nodes, sockets, and cpu cores.

It's important to note that you will achieve this if you simply run twice as many processes as you have CPU cores, since they will tend to evenly distribute over cores to share load.

I'd try something like the following, though the manual is somewhat ambiguous and I'm not 100% sure it will behave as intended, as long as you have a dual core:

<pre><code>mpirun -np 4 -npersocket 4 ./application
</code></pre>

blocks|key|2024562|text|orterun+(+Open+MPI+SPMD/MPMD启动器；mpirun/mpiexec只是指向它的符号链接)对进程绑定提供了一些支持，但它还不够灵活，不能让您在每个内核上绑定两个进程。您可以尝试使用-bycore+-bind-to-core，但当所有内核都已分配了一个进程时，它将出错。|type|unstyled|depth|inlineStyleRanges|offset|length|style|CODE|entityRanges|data|2024563|但是有一个变通的办法--你可以使用rankfile来显式地指定绑定每个等级的插槽。下面是一个示例:为了在双核CPU上运行4个进程，每个内核有2个进程，您可以执行以下操作：|2024564|mpiexec+-np+4+-H+localhost+-rf+rankfile+./application|code-block|syntax|javascript|2024565|其中，rankfile是包含以下内容的文本文件：|2024566|rank+0=localhost+slot=0:0
rank+1=localhost+slot=0:0
rank+2=localhost+slot=0:1
rank+3=localhost+slot=0:1|2024567|这将把等级0和1放在处理器0的核心0上，将等级2和3放在处理器0的核心1上。丑陋但有效：|2024568|$+mpiexec+-np+4+-H+localhost+-rf+rankfile+-tag-output+cat+/proc/self/status+%7C+grep+Cpus_allowed_list
[1,0]<stdout>:Cpus_allowed_list:+++++0
[1,1]<stdout>:Cpus_allowed_list:+++++0
[1,2]<stdout>:Cpus_allowed_list:+++++1
[1,3]<stdout>:Cpus_allowed_list:+++++1|2024569|编辑:来自你的other+question的变得很清楚，你实际上是在一个超线程的other+question上运行。然后，您必须确定逻辑处理器的物理编号(这有点令人困惑，但物理编号对应于/proc/cpuinfo中报告的processor:的值)。获得它的最简单方法是安装hwloc库。它提供了可以使用的hwloc-ls工具，如下所示：|BOLD|2024570|$+hwloc-ls+--of+console
...
++NUMANode+L#0+(P#0+48GB)+%2B+Socket+L#0+%2B+L3+L#0+(12MB)
++++L2+L#0+(256KB)+%2B+L1+L#0+(32KB)+%2B+Core+L#0
++++++PU+L#0+(P#0)++++<--+Physical+ID+0
++++++PU+L#1+(P#12)+++<--+Physical+ID+12
...|2024571|物理in列在括号中的P#之后。在您的8核情况下，第一个核心(核心0)的第二个超线程很可能具有ID+8，因此您的rankfile将如下所示：|2024572|rank+0=localhost+slot=p0
rank+1=localhost+slot=p8
rank+2=localhost+slot=p1
rank+3=localhost+slot=p9|2024573|(注意p前缀-不要省略它)|2024574|如果你没有hwloc或者你不能安装它，那么你必须自己解析/proc/cpuinfo。超线程将具有相同的physical+id和core+id值，但具有不同的processor和apicid。物理ID等于processor的值。|2024575|entityMap|0|LINK|mutability|MUTABLE|url|https://stackoverflow.com/questions/11747036/how-to-specify-the-number-of-logical-process|1^0|0|7|W|E|2U|L|0|0|0|3|8|0|0|0|0|0|M|2L|D|32|A|3S|5|48|8|7|E|0|14|E|1|0|0|A|2|1D|1|0|0|3|1|0|5|5|S|D|1F|B|1R|7|26|9|2G|6|2T|9|0^^$0|@$1|2|3|4|5|6|7|1G|8|@$9|1H|A|1I|B|C]|$9|1J|A|1K|B|C]|$9|1L|A|1M|B|C]]|D|@]|E|$]]|$1|F|3|G|5|6|7|1N|8|@]|D|@]|E|$]]|$1|H|3|I|5|J|7|1O|8|@]|D|@]|E|$K|L]]|$1|M|3|N|5|6|7|1P|8|@$9|1Q|A|1R|B|C]]|D|@]|E|$]]|$1|O|3|P|5|J|7|1S|8|@]|D|@]|E|$K|L]]|$1|Q|3|R|5|6|7|1T|8|@]|D|@]|E|$]]|$1|S|3|T|5|J|7|1U|8|@]|D|@]|E|$K|L]]|$1|U|3|V|5|6|7|1V|8|@$9|1W|A|1X|B|W]|$9|1Y|A|1Z|B|C]|$9|20|A|21|B|C]|$9|22|A|23|B|C]|$9|24|A|25|B|C]]|D|@$9|26|A|27|1|28]|$9|29|A|2A|1|2B]]|E|$]]|$1|X|3|Y|5|J|7|2C|8|@]|D|@]|E|$K|L]]|$1|Z|3|10|5|6|7|2D|8|@$9|2E|A|2F|B|C]|$9|2G|A|2H|B|C]]|D|@]|E|$]]|$1|11|3|12|5|J|7|2I|8|@]|D|@]|E|$K|L]]|$1|13|3|14|5|6|7|2J|8|@$9|2K|A|2L|B|C]]|D|@]|E|$]]|$1|15|3|16|5|6|7|2M|8|@$9|2N|A|2O|B|C]|$9|2P|A|2Q|B|C]|$9|2R|A|2S|B|C]|$9|2T|A|2U|B|C]|$9|2V|A|2W|B|C]|$9|2X|A|2Y|B|C]|$9|2Z|A|30|B|C]]|D|@]|E|$]]|$1|17|3|-4|5|6|7|31|8|@]|D|@]|E|$]]]|18|$19|$5|1A|1B|1C|E|$1D|1E]]|1F|$5|1A|1B|1C|E|$1D|1E]]]]

<code>orterun</code> (the Open MPI SPMD/MPMD launcher; <code>mpirun/mpiexec</code> are just symlinks to it) has some support for process binding but it is not flexible enough to allow you to bind two processes per core. You can try with <code>-bycore -bind-to-core</code> but it will err when all cores already have one process assigned to them.

But there is a workaround - you can use a rankfile where you explicitly specify which slot to bind each rank to. Here is an example: in order to run 4 processes on a dual-core CPU with 2 processes per core, you would do the following:

<pre><code>mpiexec -np 4 -H localhost -rf rankfile ./application
</code></pre>

where <code>rankfile</code> is a text file with the following content:

<pre><code>rank 0=localhost slot=0:0
rank 1=localhost slot=0:0
rank 2=localhost slot=0:1
rank 3=localhost slot=0:1
</code></pre>

This will place ranks 0 and 1 on core 0 of processor 0 and ranks 2 and 3 on core 1 of processor 0. Ugly but works:

<pre><code>$ mpiexec -np 4 -H localhost -rf rankfile -tag-output cat /proc/self/status | grep Cpus_allowed_list
[1,0]&lt;stdout&gt;:Cpus_allowed_list: 0
[1,1]&lt;stdout&gt;:Cpus_allowed_list: 0
[1,2]&lt;stdout&gt;:Cpus_allowed_list: 1
[1,3]&lt;stdout&gt;:Cpus_allowed_list: 1
</code></pre>

Edit: From your <a href="https://stackoverflow.com/questions/11747036/how-to-specify-the-number-of-logical-process">other question</a> is becomes clear that you are actually running on a hyperthreaded CPU. Then you would have to figure out the physical numbering of your logical processors (it's a bit confusing but physical numbering corresponds to the value of <code>processor:</code> as reported in <code>/proc/cpuinfo</code>). The easiest way to obtain it is to install the <code>hwloc</code> library. It provides the <code>hwloc-ls</code> tool that you can use like this:

<pre><code>$ hwloc-ls --of console
...
 NUMANode L#0 (P#0 48GB) + Socket L#0 + L3 L#0 (12MB)
 L2 L#0 (256KB) + L1 L#0 (32KB) + Core L#0
 PU L#0 (P#0) &lt;-- Physical ID 0
 PU L#1 (P#12) &lt;-- Physical ID 12
...
</code></pre>

Physical IDs are listed after <code>P#</code> in the brackets. In your 8-core case the second hyperthread of the first core (core 0) would most likely have ID <code>8</code> and hence your rankfile would look something like:

<pre><code>rank 0=localhost slot=p0
rank 1=localhost slot=p8
rank 2=localhost slot=p1
rank 3=localhost slot=p9
</code></pre>

(note the <code>p</code> prefix - don't omit it)

If you don't have <code>hwloc</code> or you cannot install it, then you would have to parse <code>/proc/cpuinfo</code> on your own. Hyperthreads would have the same values of <code>physical id</code> and <code>core id</code> but different <code>processor</code> and <code>apicid</code>. The physical ID is equal to the value of <code>processor</code>.

blocks|key|3660761|text|如果你使用PBS，或者类似的东西，我建议你这样提交：|type|unstyled|depth|inlineStyleRanges|entityRanges|data|3660762|qsub+-l+select=128:ncpus=40:mpiprocs=16+-v+NPROC=2048./pbs_script.csh|3660763|在本文中，我选择了128个计算节点，它们有40个核心，并使用了其中的16个。在我的例子中，每个节点有20个物理核心。|3660764|在这个提交中，我阻塞了节点的所有40个核心，没有人可以使用这些资源。它可以避免其他人使用相同的节点并与您的工作竞争。|3660765|entityMap^0|0|0|0|0^^$0|@$1|2|3|4|5|6|7|J|8|@]|9|@]|A|$]]|$1|B|3|C|5|6|7|K|8|@]|9|@]|A|$]]|$1|D|3|E|5|6|7|L|8|@]|9|@]|A|$]]|$1|F|3|G|5|6|7|M|8|@]|9|@]|A|$]]|$1|H|3|-4|5|6|7|N|8|@]|9|@]|A|$]]]|I|$]]

If you use PBS, or something like that, i would suggest this kind of submission:

qsub -l select=128:ncpus=40:mpiprocs=16 -v NPROC=2048./pbs_script.csh

In the present submission i select 128 computational nodes, that have 40 cores, and use 16 of them. In my case, i have 20 physical cores per node.

In this submission i block all the 40 cores of the node and nobody can use these resources. it can avoid other peoples from using the same node and competing with your job.

blocks|key|3750468|text|使用Open+MPI+4.0时，可以使用以下两个命令：|type|unstyled|depth|inlineStyleRanges|entityRanges|data|3750469|mpirun+--oversubscribe+-c+8+./a.out|offset|length|style|CODE|3750470|和|3750471|mpirun+-map-by+hwthread:OVERSUBSCRIBE+-c+8+./a.out|3750472|对我来说很有效(我的Ryzen+5处理器有4个核心和8个逻辑核心)。|3750473|我用一个包含实数运算的do循环进行了测试。所有的逻辑线程都被使用了，尽管看起来没有加速的好处，因为与使用-c+4选项(没有超额订阅)相比，计算花费了两倍的时间。|3750474|entityMap^0|0|0|Z|0|0|0|1E|0|0|1G|4|0^^$0|@$1|2|3|4|5|6|7|R|8|@]|9|@]|A|$]]|$1|B|3|C|5|6|7|S|8|@$D|T|E|U|F|G]]|9|@]|A|$]]|$1|H|3|I|5|6|7|V|8|@]|9|@]|A|$]]|$1|J|3|K|5|6|7|W|8|@$D|X|E|Y|F|G]]|9|@]|A|$]]|$1|L|3|M|5|6|7|Z|8|@]|9|@]|A|$]]|$1|N|3|O|5|6|7|10|8|@$D|11|E|12|F|G]]|9|@]|A|$]]|$1|P|3|-4|5|6|7|13|8|@]|9|@]|A|$]]]|Q|$]]

Using Open MPI 4.0, the two commands: 

<code>mpirun --oversubscribe -c 8 ./a.out</code>

and 

<code>mpirun -map-by hwthread:OVERSUBSCRIBE -c 8 ./a.out</code>

worked for me (I have a Ryzen 5 processor with 4 cores and 8 logical cores). 

I tested with a do loop that includes operations on real numbers. All logical threads are used, though it seems that there is no speedup benefit since computation takes double the amount of time compared to using <code>-c 4</code> option (with no oversubscribing).

blocks|key|3750500|text|您可以运行mpirun+--use-hwthread-cpus+./application|type|unstyled|depth|inlineStyleRanges|offset|length|style|CODE|entityRanges|data|3750501|在这种情况下，Open+MPI将认为处理器是由超线程提供的线程。这与它认为处理器是CPU核心时的默认行为形成对比。|3750502|使用此选项时，Open+MPI将超线程提供的线程表示为“硬件线程”，并为每个“硬件线程”分配一个Open+MPI处理器。|3750503|entityMap^0|5|14|0|0|0^^$0|@$1|2|3|4|5|6|7|L|8|@$9|M|A|N|B|C]]|D|@]|E|$]]|$1|F|3|G|5|6|7|O|8|@]|D|@]|E|$]]|$1|H|3|I|5|6|7|P|8|@]|D|@]|E|$]]|$1|J|3|-4|5|6|7|Q|8|@]|D|@]|E|$]]]|K|$]]

You can run
<code>mpirun --use-hwthread-cpus ./application</code>
In this case, Open MPI will consider that a processor is a thread provided by the Hyperthreading. This contrasts with the default behavior when it considers that a processor is a CPU core.
Open MPI denotes the threads provided by the Hyperthreading as &quot;hardware threads&quot; when you use this option, and allocates one Open MPI processor per &quot;hardware thread&quot;.

How do I assign 2 MPI processes per core?

For example, if I do <code>mpirun -np 4 ./application</code> then it should use 2 physical cores to run 4 MPI processes (2 processes per core). I am using Open MPI 1.6. I did <code>mpirun -np 4 -nc 2 ./application</code> but wasn't able to run it.

It complains <code>mpirun was unable to launch the specified application as it could not find an executable:</code>

assign two MPI processes per core

翻译质量差，导致语言生硬或混乱。

没有提供实际的解决方法或示例。

解答不清晰，无法理解或解决问题。

页面排版不美观，阅读体验差。

文章

问答

视频

学习中心

腾讯云实验室

直播

竞赛

腾讯云代码分析专区

腾讯iOA零信任安全管理系统专区

腾讯云架构师技术同盟交流圈

腾讯云数据库专区

腾讯云顾问专区

腾讯云原生专区

腾讯混元专区

腾讯云TCE专区

腾讯云Lighthouse专区

腾讯云HAI专区

腾讯云Edgeone专区

腾讯云存储专区

腾讯云智能专区

腾讯轻联专区 

腾讯云开发专区

TAPD专区

腾讯轻量云游戏服专区

腾讯云最具价值专家

腾讯云架构师技术同盟

腾讯云创作之星

腾讯云开发者先锋

腾讯云代码助手

云原生构建

TAPD 敏捷项目管理

Cloud Studio

SDK中心

API中心

命令行工具

涵盖代码开发、场景应用、自动测试全流程，助你从零构建专属AI助手

一站式MCP教程库，解锁AI应用新玩法

如何为每个内核分配2个MPI进程？例如，如果我使用mpirun -np 4 ./application，那么它应该使用2个物理内核来运行4个MPI进程(每个内核2个进程)。我使用的是Open MPI 1.6。我做了mpirun -np 4 -nc 2 ./application，但不能运行它。它抱怨mpirun was...

问为每个内核分配两个MPI进程
EN

Stack Overflow用户

社区

活动

圈层

关于

腾讯云开发者

热门产品

热门推荐

更多推荐

问为每个内核分配两个MPI进程EN