<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <p>A teljes thread itt érhető el:
      <a class="moz-txt-link-freetext" href="http://mails.dpdk.org/archives/users/2019-January/003877.html">http://mails.dpdk.org/archives/users/2019-January/003877.html</a></p>
    <p>Üdv.</p>
    <p>S<br>
    </p>
    <div class="moz-cite-prefix">2019.01.18. 14:18 keltezéssel, Sándor
      Laki írta:<br>
    </div>
    <blockquote type="cite"
      cite="mid:355909c9-c960-1ed6-fcac-7a12e06b0c0e@elte.hu">
      <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
      <p>Ez relevánsnak tűnik a crypto szempontjából és érdemes
        figyelembe venni a szívások elkerülése végett.</p>
      <p>Üdv.<br>
        S<br>
      </p>
      <div class="moz-forward-container"><br>
        <br>
        -------- Továbbított üzenet --------
        <table class="moz-email-headers-table" cellspacing="0"
          cellpadding="0" border="0">
          <tbody>
            <tr>
              <th valign="BASELINE" nowrap="nowrap" align="RIGHT">Tárgy:
              </th>
              <td>Re: [dpdk-users] Run-to-completion or Pipe-line for
                QAT PMD in DPDK</td>
            </tr>
            <tr>
              <th valign="BASELINE" nowrap="nowrap" align="RIGHT">Dátum:
              </th>
              <td>Fri, 18 Jan 2019 13:13:43 +0000</td>
            </tr>
            <tr>
              <th valign="BASELINE" nowrap="nowrap" align="RIGHT">Feladó:
              </th>
              <td>Trahe, Fiona <a class="moz-txt-link-rfc2396E"
                  href="mailto:fiona.trahe@intel.com"
                  moz-do-not-send="true"><fiona.trahe@intel.com></a></td>
            </tr>
            <tr>
              <th valign="BASELINE" nowrap="nowrap" align="RIGHT">Címzett:
              </th>
              <td>Changchun Zhang <a class="moz-txt-link-rfc2396E"
                  href="mailto:changchun.zhang@oracle.com"
                  moz-do-not-send="true"><changchun.zhang@oracle.com></a>,
                <a class="moz-txt-link-abbreviated"
                  href="mailto:users@dpdk.org" moz-do-not-send="true">users@dpdk.org</a>
                <a class="moz-txt-link-rfc2396E"
                  href="mailto:users@dpdk.org" moz-do-not-send="true"><users@dpdk.org></a></td>
            </tr>
            <tr>
              <th valign="BASELINE" nowrap="nowrap" align="RIGHT">CC: </th>
              <td>Trahe, Fiona <a class="moz-txt-link-rfc2396E"
                  href="mailto:fiona.trahe@intel.com"
                  moz-do-not-send="true"><fiona.trahe@intel.com></a></td>
            </tr>
          </tbody>
        </table>
        <br>
        <br>
        Hi Alex,<br>
        <br>
        <blockquote type="cite">-----Original Message-----<br>
          From: users [<a class="moz-txt-link-freetext"
            href="mailto:users-bounces@dpdk.org" moz-do-not-send="true">mailto:users-bounces@dpdk.org</a>]
          On Behalf Of Changchun Zhang<br>
          Sent: Thursday, January 17, 2019 11:01 PM<br>
          To: <a class="moz-txt-link-abbreviated"
            href="mailto:users@dpdk.org" moz-do-not-send="true">users@dpdk.org</a><br>
          Subject: [dpdk-users] Run-to-completion or Pipe-line for QAT
          PMD in DPDK<br>
          <br>
          Hi,<br>
          <br>
          <br>
          <br>
          I have user question on using the QAT device in the DPDK.<br>
          <br>
          In the real design, after calling enqueuer_burst() on the
          specified queue pair at one of the lcore,<br>
          usually which one is usually done?<br>
          <br>
          1. should we do run-to-completion to call dequeuer_burst()
          waiting for the device finishing the<br>
          crypto operation,<br>
          <br>
          2. or should we do pipe-line, in which we return right after
          enqueuer_burst() and release the CPU.<br>
          And call dequeuer_burst() on other thread function?<br>
          <br>
          Option 1 is more like synchronous and can be seen on all the
          DPDK crypto examples, while option 2 is<br>
          asynchronous which I have never seen in any reference design
          if I missed anything.<br>
        </blockquote>
        [Fiona] Option 2 is not possible with QAT - the dequeue must be
        called in the same thread as the enqueue. This is<br>
        optimised without atomics for best performance - if this is a
        problem let us know. However best performance is not quite using
        option 1 and not a synchronous blocking method. If you enqueue
        and then go straight to dequeue, you're not getting the best
        advantage from the<br>
        cycles freed up by offloading. i.e. best to enqueue a burst,
        then go do some other work, like maybe collecting more requests
        for next enqueue or other processing, then dequeue. Take and
        process whatever ops are dequeued - this<br>
        will not necessarily match up with the number you've enqueued -
        depends on how quickly you call the dequeue.<br>
        Don't wait until all the enqueued ops are dequeued before
        enqueuing the next batch.<br>
        SO it's asynchronous. But in the same thread.<br>
        You'll get best throughput when you keep the input filled up so
        the device has operations to work on and<br>
        regularly dequeue a burst. Dequeuing too often will waste cycles
        in the overhead calling the API, dequeuing too<br>
        slowly will cause the device to back up. Ideally tune for your
        application to find the sweet spot in<br>
        between these 2 extremes. </div>
      <div id="DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2"><br>
        <table style="border-top: 1px solid #D3D4DE;">
          <tbody>
            <tr>
              <td style="width: 55px; padding-top: 18px;"><a
href="https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient"
                  target="_blank" moz-do-not-send="true"><img
src="https://ipmcdn.avast.com/images/icons/icon-envelope-tick-round-orange-animated-no-repeat-v1.gif"
                    alt="" style="width: 46px; height: 29px;"
                    moz-do-not-send="true" width="46" height="29"></a></td>
              <td style="width: 470px; padding-top: 17px; color:
                #41424e; font-size: 13px; font-family: Arial, Helvetica,
                sans-serif; line-height: 18px;">Mentes a vírusoktól. <a
href="https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient"
                  target="_blank" style="color: #4453ea;"
                  moz-do-not-send="true">www.avast.com</a> </td>
            </tr>
          </tbody>
        </table>
        <a href="#DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2" width="1"
          height="1" moz-do-not-send="true"> </a></div>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <pre class="moz-quote-pre" wrap="">_______________________________________________
P4 mailing list
<a class="moz-txt-link-abbreviated" href="mailto:P4@plc.inf.elte.hu">P4@plc.inf.elte.hu</a>
<a class="moz-txt-link-freetext" href="https://plc.inf.elte.hu/mailman/listinfo/p4">https://plc.inf.elte.hu/mailman/listinfo/p4</a>
</pre>
    </blockquote>
    <pre class="moz-signature" cols="72">-- 
Sándor Laki, PhD
Assistant professor
Department of Information Systems
Eötvös Loránd University
Pázmány Péter stny. 1/C
H-1117, Budapest, Hungary
Room 2.506
Web: <a class="moz-txt-link-freetext" href="http://lakis.web.elte.hu">http://lakis.web.elte.hu</a>
Phone: +36 1 372 2869 / 8477
Cell: +36 70 374 2646</pre>
  </body>
</html>