<html>
  <head>
    <meta content="text/html; charset=ISO-8859-1"
      http-equiv="Content-Type">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    <div class="moz-cite-prefix">I tried downgrading to 3.2 and the rdma
      connection works, with substantially improved performance in some
      cases over 3.3 using IP over IB.<br>
      <br>
      Is there a version of 3.3 that anyone is using in production with
      rdma that I can try?<br>
      <br>
      Iain<br>
      <br>
      <br>
      On 05/30/13 09:24, Iain Buchanan wrote:<br>
    </div>
    <blockquote cite="mid:51A70CD6.2030708@gmail.com" type="cite">
      <meta content="text/html; charset=ISO-8859-1"
        http-equiv="Content-Type">
      <div class="moz-cite-prefix">I'm now able to connect to volumes
        using rdma when they are created using "tcp,rdma", but it
        appears that the data is still being transferred over ethernet.&nbsp;
        If I run "ifstat" while running iozone I can see a lot of data
        being moved through the ethernet adapter (nothing else is on the
        box) and the performance is basically identical to plain
        ethernet.&nbsp; When I create the volume with just "rdma" I can't
        even mount the volume again (see error below).<br>
        <br>
        I've modified all the volume files, adding the
        transport.rdma.listen-port lines after completing set-up
        (anywhere there is a "option transport-type rdma" I've added
        this line - quite a few places).<br>
        <br>
        &nbsp;&nbsp;&nbsp; volume create storage transport tcp,rdma
        my_server1:/data/area<br>
        &nbsp;&nbsp;&nbsp; volume add-brick storage replica 2 my_server2:/data/area<br>
        &nbsp;&nbsp;&nbsp; volume start storage<br>
        &nbsp;&nbsp;&nbsp; mount -t glusterfs my_server:storage.rdma /mnt/storage<br>
        <br>
        I'm now back to the error I had before the workaround on
        tcp,rdma - the lines are in the config files, and I've restarted
        the service.&nbsp; I can see my alterations in the config dumped into
        the log, but then it returns to the original errors:<br>
        <br>
        [2013-05-30 09:13:41.871408] E [rdma.c:4604:tcp_connect_finish]
        0-storage-client-0: tcp connect to&nbsp; failed (Connection refused)<br>
        [2013-05-30 09:13:41.871467] W [rdma.c:4187:gf_rdma_disconnect]
        (--&gt;/usr/sbin/glusterfs(main+0x34d) [0x7f563a24c3ed]
        (--&gt;/usr/lib/libglusterfs.so.0(+0x3bd17) [0x7f5639de5d17]
        (--&gt;/usr/lib/glusterfs/3.3.1/rpc-transport/rdma.so(+0x5231)
        [0x7f5634012231]))) 0-storage-client-0: disconnect called
        (peer:)<br>
        <br>
        I saw a message on the mailing list that seems to suggest plain
        "rdma" doesn't work in the 3.3 series - is this correct?<br>
        <a moz-do-not-send="true" class="moz-txt-link-freetext"
href="http://www.gluster.org/pipermail/gluster-users/2013-January/035115.html">http://www.gluster.org/pipermail/gluster-users/2013-January/035115.html</a><br>
        <br>
        (I can run ib_read_bw etc. between the two servers without
        problems.)<br>
        <br>
        Iain<br>
        <br>
        <br>
        On 05/30/13 08:11, Iain Buchanan wrote:<br>
      </div>
      <blockquote cite="mid:51A6FBB2.5010007@gmail.com" type="cite">
        <meta content="text/html; charset=ISO-8859-1"
          http-equiv="Content-Type">
        <div class="moz-cite-prefix">Just to confirm - putting the line
          "option transport.rdma.listen-port 24008" into the first
          "volume" block in the two files with "rdma" in their names
          under /var/lib/glusterd/vols/&lt;volumename&gt; seems to have
          fixed the issue.&nbsp; I'm now able to mount and I can run iozone
          on a single node.&nbsp; I'll give it a go with two nodes and see if
          rdma makes any difference.<br>
          <br>
          Thanks for your help Joe!<br>
          <br>
          Iain<br>
          <br>
          <br>
          On 05/30/13 07:59, Joe Julian wrote:<br>
        </div>
        <blockquote cite="mid:51A6F8DD.9030705@julianfamily.org"
          type="cite">
          <meta content="text/html; charset=ISO-8859-1"
            http-equiv="Content-Type">
          <div class="moz-cite-prefix">On 05/29/2013 11:42 PM, Iain
            Buchanan wrote:<br>
          </div>
          <blockquote cite="mid:51A6F4DB.5050704@gmail.com" type="cite">
            <meta content="text/html; charset=ISO-8859-1"
              http-equiv="Content-Type">
            <div class="moz-cite-prefix">Thanks Joe,<br>
              <br>
              I tried mounting with the ".rdma" suffix after creating
              using "transport tcp,rdma" and I get the same "tcp connect
              to&nbsp; failed" messages in the log - I'm using the version
              from the semiosis PPA at <a moz-do-not-send="true"
                class="moz-txt-link-freetext"
                href="https://launchpad.net/%7Esemiosis/+archive/ubuntu-glusterfs-3.3">https://launchpad.net/~semiosis/+archive/ubuntu-glusterfs-3.3</a><br>
              <br>
              Looking at the dates on there I don't think it includes
              this fix.&nbsp; I've sent the maintainer a message asking if
              they could update it.&nbsp; In the meantime would Niels de Vos'
              workaround work?&nbsp; (Setting transport.rdma.listen-port to
              24008 in the glusterd.vol?)<br>
            </div>
          </blockquote>
          <br>
          Yes, I had someone else try that and it worked. Remember, any
          changes made to the volume through the CLI will reset that
          configuration, causing it to stop working until you edit
          again.<br>
          <br>
          <blockquote cite="mid:51A6F4DB.5050704@gmail.com" type="cite">
            <div class="moz-cite-prefix"> <br>
              Iain<br>
              <br>
              On 05/30/13 07:06, Joe Julian wrote:<br>
            </div>
            <blockquote cite="mid:51A6EC58.3070701@julianfamily.org"
              type="cite">
              <meta content="text/html; charset=ISO-8859-1"
                http-equiv="Content-Type">
              <div class="moz-cite-prefix">On 05/29/2013 10:33 PM, Iain
                Buchanan wrote:<br>
              </div>
              <blockquote
                cite="mid:B4A78FEC-456B-46A8-B876-69F4727B1DB2@gmail.com"
                type="cite">
                <meta http-equiv="Content-Type" content="text/html;
                  charset=ISO-8859-1">
                Hi,
                <div><br>
                  <div>I'm running GlusterFS&nbsp;3.3.1-ubuntu1~precise9 and
                    I'm having some problems with the "rdma" and
                    "tcp,rdma" options I hope someone can help me with.
                    &nbsp;</div>
                  <div><br>
                  </div>
                  <div>1. What does "tcp,rdma" actually do - does it let
                    you mix both types of client? &nbsp;(I did a few tests
                    with iozone and found it gave identical performance
                    to the "tcp".)</div>
                  <div><br>
                  </div>
                  <div>2. I can't get "rdma" to work, even in the
                    simplest case with a single node.</div>
                  <div><font face="Courier"><span class="Apple-tab-span"
                        style="white-space:pre"> </span>volume create
                      storage transport transport rdma
                      my_server:/data/area</font></div>
                  <div><font face="Courier"><span class="Apple-tab-span"
                        style="white-space:pre"> </span>volume start
                      storage</font></div>
                  <div><font face="Courier"><span class="Apple-tab-span"
                        style="white-space:pre"> </span>mount -t
                      glusterfs my_server:storage /mnt/storage</font></div>
                  <div><br>
                  </div>
                  <div>The last line hangs. &nbsp;Looking in
                    /var/log/glusterfs I can see the log for the volume:</div>
                  <div><br>
                  </div>
                  <div>
                    <div><font face="Courier">[2013-05-30
                        06:24:19.605315] E
                        [rdma.c:4604:tcp_connect_finish]
                        0-storage-client-0: <b>tcp connect to &nbsp;failed
                          (Connection refused)</b></font></div>
                    <div><font face="Courier">[2013-05-30
                        06:24:19.605713] W
                        [rdma.c:4187:gf_rdma_disconnect]
                        (--&gt;/usr/sbin/glusterfs(main+0x34d)
                        [0x7f374d38a3ed]
                        (--&gt;/usr/lib/libglusterfs.so.0(+0x3bd17)
                        [0x7f374cf23d17]
                        (--&gt;/usr/lib/glusterfs/3.3.1/rpc-transport/rdma.so(+0x5231)
                        [0x7f3743398231]))) 0-storage-client-0:
                        disconnect called (peer:)</font></div>
                    <div><font face="Courier">[2013-05-30
                        06:24:19.605763] W
                        [rdma.c:4521:gf_rdma_handshake_pollerr]
                        (--&gt;/usr/sbin/glusterfs(main+0x34d)
                        [0x7f374d38a3ed]
                        (--&gt;/usr/lib/libglusterfs.so.0(+0x3bd17)
                        [0x7f374cf23d17]
                        (--&gt;/usr/lib/glusterfs/3.3.1/rpc-transport/rdma.so(+0x5150)
                        [0x7f3743398150]))) 0-rpc-transport/rdma:
                        storage-client-0: peer () disconnected, cleaning
                        up</font></div>
                  </div>
                  <div><br>
                  </div>
                  <div>This block repeats every few seconds - the line
                    "tcp connect to &nbsp;failed" looks like it has lost the
                    server name somehow? &nbsp;</div>
                  <div><br>
                  </div>
                  <div>Iain</div>
                </div>
                <br>
              </blockquote>
              If you've installed from the yum repo (<a
                moz-do-not-send="true" class="moz-txt-link-freetext"
                href="http://goo.gl/s077x">http://goo.gl/s077x</a>) that
              shouldn't be happening. kkeithley applied the patch. If
              not, rdma's broken in 3.3.[01].
              <meta http-equiv="content-type" content="text/html;
                charset=ISO-8859-1">
              <a moz-do-not-send="true"
                href="https://bugzilla.redhat.com/show_bug.cgi?id=849122">https://bugzilla.redhat.com/show_bug.cgi?id=849122</a><br>
              <br>
              To mount via rdma when using tcp,rdma, mount -t glusterfs
              server1:myvol.rdma /mnt/foo<br>
              <br>
              <br>
            </blockquote>
          </blockquote>
          <br>
          <br>
          <fieldset class="mimeAttachmentHeader"></fieldset>
          <br>
          <pre wrap="">_______________________________________________
Gluster-users mailing list
<a moz-do-not-send="true" class="moz-txt-link-abbreviated" href="mailto:Gluster-users@gluster.org">Gluster-users@gluster.org</a>
<a moz-do-not-send="true" class="moz-txt-link-freetext" href="http://supercolony.gluster.org/mailman/listinfo/gluster-users">http://supercolony.gluster.org/mailman/listinfo/gluster-users</a></pre>
        </blockquote>
        <br>
      </blockquote>
      <br>
    </blockquote>
    <br>
  </body>
</html>