<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xmlns:taxo="http://purl.org/rss/1.0/modules/taxonomy/" version="2.0">
  <channel>
    <title>topic Re: One Sided MPI Communication on Windows platform in Intel® MPI Library</title>
    <link>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849662#M1444</link>
    <description>&lt;DIV style="margin:0px;"&gt;Thanks Gergana. &lt;BR /&gt;&lt;BR /&gt;The release notes states that ONLY the folowing is not supported (...not the enire one-sided comm feature) - &lt;BR /&gt;"Passive target one-sided communication when the target process does not call any MPI functions", which I understand. &lt;BR /&gt;&lt;BR /&gt;As mater of fact, I verified it with the Intel MPI library (using the Sockets as comm channel), that passive sync works fine, if there is a seperate thread that has a MPI call (such as MPI_Recv ) pending, while the one sided comm is occuring. &lt;BR /&gt;&lt;BR /&gt;I am OK with the above restriction, but my question was specific to the usage of "remote DMA" feature of the Infiniband HCA, for performing one-sided communication (with passive sync), so that the target process does not have to 'actively' participate in the one-sided comm data transfer ?&lt;BR /&gt;&lt;BR /&gt;To clarify, I need the answer for Intel MPI library, for Windows, using WinOF stack.&lt;BR /&gt;&lt;BR /&gt;Sashi&lt;/DIV&gt;
&lt;BR /&gt;</description>
    <pubDate>Thu, 23 Jul 2009 00:55:06 GMT</pubDate>
    <dc:creator>sashibala</dc:creator>
    <dc:date>2009-07-23T00:55:06Z</dc:date>
    <item>
      <title>One Sided MPI Communication on Windows platform</title>
      <link>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849660#M1442</link>
      <description>I am currently evaluating the Intel MPI for Windows platform, with Infiniband channel, and using WinOF DAPL.&lt;BR /&gt;&lt;BR /&gt;I will be implementing a few 'One-Sided' communication methods to maximize data transfer efficiency, especially with image data buffer transfers (...approx 4 MBytes long), between multiple nodes.&lt;BR /&gt;&lt;BR /&gt;Could you please confirm if the Intel MPI has 'implemented' the 'one-sided' communication, with 'passive' synchronization, by taking advantage of the "remote DMA" capability of Infiniband HCA....and if so, would it work with the new WinOF ?&lt;BR /&gt;&lt;BR /&gt;Ideally, I am looking for the MVAPICH-2 type implementation, except on Windows platform. &lt;BR /&gt;Pls see &lt;A href="http://www.mcs.anl.gov/~thakur/papers/onesided-iba.pdf" target="_blank"&gt;http://www.mcs.anl.gov/~thakur/papers/onesided-iba.pdf&lt;/A&gt;&lt;BR /&gt;&lt;BR /&gt;Thanks,&lt;BR /&gt;Sashi&lt;BR /&gt;</description>
      <pubDate>Tue, 21 Jul 2009 22:05:33 GMT</pubDate>
      <guid>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849660#M1442</guid>
      <dc:creator>sashibala</dc:creator>
      <dc:date>2009-07-21T22:05:33Z</dc:date>
    </item>
    <item>
      <title>Re: One Sided MPI Communication on Windows platform</title>
      <link>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849661#M1443</link>
      <description>&lt;P&gt;Hi Sashi,&lt;/P&gt;
&lt;P&gt;In the current release of the Intel MPI Library 3.2 Update 1 for Windows*, one-sided communication is not yet supported (check out the &lt;A href="http://software.intel.com/en-us/articles/intel-mpi-library-documentation/" target="_blank"&gt;Release Notes&lt;/A&gt; for details). We've only recently had it added to the Linux product.&lt;/P&gt;
&lt;P&gt;If you feel this is an important functionality, I can go ahead and submit a feature request to our development team.&lt;/P&gt;
&lt;P&gt;Regards,&lt;BR /&gt;~Gergana&lt;/P&gt;</description>
      <pubDate>Wed, 22 Jul 2009 23:07:23 GMT</pubDate>
      <guid>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849661#M1443</guid>
      <dc:creator>Gergana_S_Intel</dc:creator>
      <dc:date>2009-07-22T23:07:23Z</dc:date>
    </item>
    <item>
      <title>Re: One Sided MPI Communication on Windows platform</title>
      <link>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849662#M1444</link>
      <description>&lt;DIV style="margin:0px;"&gt;Thanks Gergana. &lt;BR /&gt;&lt;BR /&gt;The release notes states that ONLY the folowing is not supported (...not the enire one-sided comm feature) - &lt;BR /&gt;"Passive target one-sided communication when the target process does not call any MPI functions", which I understand. &lt;BR /&gt;&lt;BR /&gt;As mater of fact, I verified it with the Intel MPI library (using the Sockets as comm channel), that passive sync works fine, if there is a seperate thread that has a MPI call (such as MPI_Recv ) pending, while the one sided comm is occuring. &lt;BR /&gt;&lt;BR /&gt;I am OK with the above restriction, but my question was specific to the usage of "remote DMA" feature of the Infiniband HCA, for performing one-sided communication (with passive sync), so that the target process does not have to 'actively' participate in the one-sided comm data transfer ?&lt;BR /&gt;&lt;BR /&gt;To clarify, I need the answer for Intel MPI library, for Windows, using WinOF stack.&lt;BR /&gt;&lt;BR /&gt;Sashi&lt;/DIV&gt;
&lt;BR /&gt;</description>
      <pubDate>Thu, 23 Jul 2009 00:55:06 GMT</pubDate>
      <guid>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849662#M1444</guid>
      <dc:creator>sashibala</dc:creator>
      <dc:date>2009-07-23T00:55:06Z</dc:date>
    </item>
    <item>
      <title>Re: One Sided MPI Communication on Windows platform</title>
      <link>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849663#M1445</link>
      <description>&lt;BR /&gt;Hi,&lt;BR /&gt;Support for one sided operations should be at similar level for all supported fabrics. Could you submit an issue report at &lt;A href="https://premier.intel.com"&gt;https://premier.intel.com&lt;/A&gt;? Please provide versions of used software and instructions how to reproduce an issue in the report. So we will be able to investigate it.&lt;BR /&gt;&lt;BR /&gt;Best regards,&lt;BR /&gt;Andrey</description>
      <pubDate>Thu, 23 Jul 2009 07:08:37 GMT</pubDate>
      <guid>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849663#M1445</guid>
      <dc:creator>Andrey_D_Intel</dc:creator>
      <dc:date>2009-07-23T07:08:37Z</dc:date>
    </item>
    <item>
      <title>Re: One Sided MPI Communication on Windows platform</title>
      <link>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849664#M1446</link>
      <description>&lt;DIV style="margin:0px;"&gt;&lt;/DIV&gt;
Andrey:&lt;BR /&gt;Actually I don't have an application issue at this time.It'sonly a question of how Intel (as an 'implemetor')chose to 'implement' the one-sided communication spec of MPI-2. &lt;BR /&gt;&lt;BR /&gt;As mentioned before, the MVAPICH-2 implementation (of MPI-2), takes advantage of the remote DMA capability of the IB hardware to improve the through-putpeformance significantly, on one-sided comm.&lt;BR /&gt;&lt;BR /&gt;Thanks,&lt;BR /&gt;Sashi&lt;BR /&gt;</description>
      <pubDate>Thu, 23 Jul 2009 16:37:16 GMT</pubDate>
      <guid>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849664#M1446</guid>
      <dc:creator>sashibala</dc:creator>
      <dc:date>2009-07-23T16:37:16Z</dc:date>
    </item>
    <item>
      <title>Re: One Sided MPI Communication on Windows platform</title>
      <link>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849665#M1447</link>
      <description>&lt;DIV style="margin:0px;"&gt;
&lt;DIV id="quote_reply" style="width: 100%; margin-top: 5px;"&gt;
&lt;DIV style="margin-left:2px;margin-right:2px;"&gt;Quoting - &lt;A href="https://community.intel.com/en-us/profile/435393"&gt;sashibala&lt;/A&gt;&lt;/DIV&gt;
&lt;DIV style="background-color:#E5E5E5; padding:5px;border: 1px; border-style: inset;margin-left:2px;margin-right:2px;"&gt;&lt;EM&gt; Andrey:&lt;BR /&gt;Actually I don't have an application issue at this time.It'sonly a question of how Intel (as an 'implemetor')chose to 'implement' the one-sided communication spec of MPI-2. &lt;BR /&gt;&lt;BR /&gt;As mentioned before, the MVAPICH-2 implementation (of MPI-2), takes advantage of the remote DMA capability of the IB hardware to improve the through-putpeformance significantly, on one-sided comm.&lt;BR /&gt;&lt;BR /&gt;Thanks,&lt;BR /&gt;Sashi&lt;BR /&gt;&lt;/EM&gt;&lt;/DIV&gt;
&lt;/DIV&gt;
&lt;/DIV&gt;
&lt;BR /&gt;Hi Sashi,&lt;BR /&gt;&lt;BR /&gt;Intel MPI Library doesn't support passive one-sided operations because we've never seen real customers who use this feature. It might be interesting for academic purposes but there is no real applications so far.&lt;BR /&gt;&lt;BR /&gt;Regards!&lt;BR /&gt; Dmitry&lt;BR /&gt;&lt;BR /&gt;</description>
      <pubDate>Tue, 28 Jul 2009 12:04:07 GMT</pubDate>
      <guid>https://community.intel.com/t5/Intel-MPI-Library/One-Sided-MPI-Communication-on-Windows-platform/m-p/849665#M1447</guid>
      <dc:creator>Dmitry_K_Intel2</dc:creator>
      <dc:date>2009-07-28T12:04:07Z</dc:date>
    </item>
  </channel>
</rss>

