<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html;
      charset=windows-1252">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <p>Hi Yuhao,<br>
      <br>
      Since ram is relatively inexpensive, if you have another 64GB
      laying around, why don't you stick them in to make it total 128GB
      to serve your 482TB?<br>
      <br>
      From what I have read it appears there is a general recommendation
      1GB Ram / 1 TB Disk<br>
    </p>
    Our setup here we are using a Raid card to setup RAID 10 and present
    it to gluster as a single brick. I have seen issue if i directly
    write to the RAID 10, I would get around 250MB/sec, via gluster
    mounted volume 100MB and if i put NFS on top it would reduce to
    30MB/sec<br>
    <br>
    Regards,<br>
    Edy<br>
    <br>
    <div class="moz-cite-prefix">On 8/8/2018 1:49 PM, Yuhao Zhang wrote:<br>
    </div>
    <blockquote type="cite"
      cite="mid:AEE92B0D-ED09-45DD-BE6E-B254EBF8BD8B@gmail.com">
      <meta http-equiv="Content-Type" content="text/html;
        charset=windows-1252">
      Hi Xavi,
      <div class=""><br class="">
      </div>
      <div class="">Thank you for the suggestions, these are extremely
        helpful. I haven't thought it could be ZFS problem. I went back
        and checked a longer monitoring window and now I can see a
        pattern. Please see this attached Grafana screenshot (also
        available here:�<a href="https://cl.ly/070J2y3n1u0F" class=""
          moz-do-not-send="true">https://cl.ly/070J2y3n1u0F</a>�. Note
        that the data gaps were when I took down the server for
        rebooting):</div>
      <div class=""><br class="">
      </div>
      <div class=""><img apple-inline="yes"
          id="E2C98C60-010D-41C6-A758-54A51DE54118"
          src="cid:part2.44048C42.E1F87799@edylie.net" class=""
          height="586" width="1249"><br class="">
        <div><br class="">
        </div>
        <div>Between 8/4 - 8/6, I tried two transfer tests, and
          experienced 2 the gluster hanging problems. One during the
          first transfer, and another one happened shortly after the
          second transfer. I blocked both in pink lines.�</div>
        <div><br class="">
        </div>
        <div>Looks like during my transfer tests, free memory was almost
          exhausted. The system has a very high cached memory, which I
          think was due to ZFS ARC. However, I am under the impression
          that ZFS will release space from ARC if it observes low system
          available memory. I am not sure why it didn't do that.�</div>
        <div><br class="">
        </div>
        <div>I did't tweak related ZFS parameters.�zfs_arc_max was set
          to 0 (default value). According to doc, it is "Max arc size of
          ARC in bytes. If set to 0 then it will consume 1/2 �of �system
          RAM." So it appeared that this setting didn't work.</div>
        <div><br class="">
        </div>
        <div>When the server was under heavy IO, the used memory was
          instead decreased, which I can't explain.</div>
        <div><br class="">
        </div>
        <div>May I ask if you, or anyone else in this group, has
          recommendation on ZFS settings for my setup? My server has
          64GB physical memory and 150GB SSD space reserved for
          L2_ARC.The zpool has 6 vdevs and each has 12TB * 10 hard
          drives on raidz2. Total usable space in the zpool is 482TB.</div>
        <div><br class="">
        </div>
        <div>Thank you,</div>
        <div>Yuhao</div>
        <div><br class="">
        </div>
        <div>
          <blockquote type="cite" class="">
            <div class="">On Aug 7, 2018, at 01:36, Xavi Hernandez &lt;<a
                href="mailto:jahernan@redhat.com" class=""
                moz-do-not-send="true">jahernan@redhat.com</a>&gt;
              wrote:</div>
            <br class="Apple-interchange-newline">
            <div class="">
              <div dir="auto" class="">
                <div class="">Hi Yuhao,�<br class="">
                  <br class="">
                  <div class="gmail_quote">
                    <div dir="ltr" class="">On Mon, 6 Aug 2018, 15:26
                      Yuhao Zhang, &lt;<a href="mailto:zzyzxd@gmail.com"
                        class="" moz-do-not-send="true">zzyzxd@gmail.com</a>&gt;
                      wrote:<br class="">
                    </div>
                    <blockquote class="gmail_quote" style="margin:0 0 0
                      .8ex;border-left:1px #ccc solid;padding-left:1ex">
                      <div
                        style="word-wrap:break-word;line-break:after-white-space"
                        class="">
                        <div class="">Hello,</div>
                        <div class=""><br class="">
                        </div>
                        I just experienced another hanging one hour ago
                        and the server was not even under heavy IO.
                        <div class=""><br class="">
                        </div>
                        <div class="">Atin, I attached the process
                          monitoring results and another statedump.</div>
                        <div class=""><br class="">
                        </div>
                        <div class="">Xavi, ZFS was fine, during the
                          hanging, I can still write directly to the ZFS
                          volume. My ZFS version: ZFS: Loaded module
                          v0.6.5.6-0ubuntu16, ZFS pool version 5000, ZFS
                          filesystem version 5</div>
                      </div>
                    </blockquote>
                  </div>
                </div>
                <div dir="auto" class=""><br class="">
                </div>
                <div dir="auto" class="">I highly recommend you to
                  upgrade to version 0.6.5.8 at least. It fixes a kernel
                  panic that can happen when used with gluster. However
                  this is not your current problem.</div>
                <div dir="auto" class=""><br class="">
                </div>
                <div dir="auto" class="">Top statistics show low
                  available memory and high CPU utilization of kswapd
                  process (along with one of the gluster processes).
                  I've seen frequent memory management problems with
                  ZFS. Have you configured any ZFS parameters? It's
                  highly recommendable to tweak some memory limits.</div>
                <div dir="auto" class=""><br class="">
                </div>
                <div dir="auto" class="">If that were the problem,
                  there's one thing that should alleviate it (and see if
                  it could be related):</div>
                <div dir="auto" class=""><br class="">
                </div>
                <div dir="auto" class="">echo 3
                  &gt;/proc/sys/vm/drop_caches</div>
                <div dir="auto" class=""><br class="">
                </div>
                <div dir="auto" class="">This should be done on all
                  bricks from time to time. You can wait until the
                  problem appears, but in this case the recovery time
                  can be larger.�</div>
                <div dir="auto" class=""><br class="">
                </div>
                <div dir="auto" class="">I think this should fix the
                  high CPU usage of kswapd. If so, we'll need to tweak
                  some ZFS parameters.</div>
                <div dir="auto" class=""><br class="">
                </div>
                <div dir="auto" class="">I'm not sure if the high CPU
                  usage of gluster could be related to this or not.</div>
                <div dir="auto" class=""><br class="">
                </div>
                <div dir="auto" class="">Xavi</div>
                <div dir="auto" class="">
                  <div class="gmail_quote">
                    <blockquote class="gmail_quote" style="margin:0 0 0
                      .8ex;border-left:1px #ccc solid;padding-left:1ex">
                      <div
                        style="word-wrap:break-word;line-break:after-white-space"
                        class="">
                        <div class="">
                          <div class=""><br class="">
                          </div>
                          <div class="">Thank you,</div>
                          <div class="">Yuhao</div>
                        </div>
                      </div>
                    </blockquote>
                  </div>
                </div>
              </div>
            </div>
          </blockquote>
        </div>
        <br class="">
      </div>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <br>
      <pre wrap="">_______________________________________________
Gluster-users mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Gluster-users@gluster.org">Gluster-users@gluster.org</a>
<a class="moz-txt-link-freetext" href="https://lists.gluster.org/mailman/listinfo/gluster-users">https://lists.gluster.org/mailman/listinfo/gluster-users</a></pre>
    </blockquote>
    <br>
  </body>
</html>