<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
  </head>
  <body>
    <p>Dear Artem,</p>
    <p>can you also provide some information w.r.t your xfs filesystem,
      i.e. xfs_info of your block device?</p>
    <p><br>
    </p>
    <p>Regards,</p>
    <p>Felix<br>
    </p>
    <div class="moz-cite-prefix">On 30/04/2020 17:27, Artem Russakovskii
      wrote:<br>
    </div>
    <blockquote type="cite"
cite="mid:CAD+dzQdT7MGWzNPvxpYfuSR=sR9HNCu5=EGepSMOuZVDGGWmAw@mail.gmail.com">
      <meta http-equiv="content-type" content="text/html; charset=UTF-8">
      <div dir="auto">
        <div>Hi Strahil, in the original email I included both the times
          for the first and subsequent reads on the fuse mounted gluster
          volume as well as the xfs filesystem the gluster data resides
          on (this is the brick, right?). </div>
        <div dir="auto"><br>
          <div class="gmail_quote" dir="auto">
            <div dir="ltr" class="gmail_attr">On Thu, Apr 30, 2020, 7:44
              AM Strahil Nikolov &lt;<a
                href="mailto:hunter86_bg@yahoo.com"
                moz-do-not-send="true">hunter86_bg@yahoo.com</a>&gt;
              wrote:<br>
            </div>
            <blockquote class="gmail_quote" style="margin:0 0 0
              .8ex;border-left:1px #ccc solid;padding-left:1ex">On April
              30, 2020 4:24:23 AM GMT+03:00, Artem Russakovskii &lt;<a
                href="mailto:archon810@gmail.com" target="_blank"
                rel="noreferrer" moz-do-not-send="true">archon810@gmail.com</a>&gt;
              wrote:<br>
              &gt;Hi all,<br>
              &gt;<br>
              &gt;We have 500GB and 10TB 4x1 replicate xfs-based gluster
              volumes, and the<br>
              &gt;10TB one especially is extremely slow to do certain
              things with (and<br>
              &gt;has<br>
              &gt;been since gluster 3.x when we started). We're
              currently on 5.13.<br>
              &gt;<br>
              &gt;The number of files isn't even what I'd consider that
              great - under<br>
              &gt;100k<br>
              &gt;per dir.<br>
              &gt;<br>
              &gt;Here are some numbers to look at:<br>
              &gt;<br>
              &gt;On gluster volume in a dir of 45k files:<br>
              &gt;The first time<br>
              &gt;<br>
              &gt;time find | wc -l<br>
              &gt;45423<br>
              &gt;real    8m44.819s<br>
              &gt;user    0m0.459s<br>
              &gt;sys     0m0.998s<br>
              &gt;<br>
              &gt;And again<br>
              &gt;<br>
              &gt;time find | wc -l<br>
              &gt;45423<br>
              &gt;real    0m34.677s<br>
              &gt;user    0m0.291s<br>
              &gt;sys     0m0.754s<br>
              &gt;<br>
              &gt;<br>
              &gt;If I run the same operation on the xfs block device
              itself:<br>
              &gt;The first time<br>
              &gt;<br>
              &gt;time find | wc -l<br>
              &gt;45423<br>
              &gt;real    0m13.514s<br>
              &gt;user    0m0.144s<br>
              &gt;sys     0m0.501s<br>
              &gt;<br>
              &gt;And again<br>
              &gt;<br>
              &gt;time find | wc -l<br>
              &gt;45423<br>
              &gt;real    0m0.197s<br>
              &gt;user    0m0.088s<br>
              &gt;sys     0m0.106s<br>
              &gt;<br>
              &gt;<br>
              &gt;I'd expect a performance difference here but just as
              it was several<br>
              &gt;years<br>
              &gt;ago when we started with gluster, it's still huge, and
              simple file<br>
              &gt;listings<br>
              &gt;are incredibly slow.<br>
              &gt;<br>
              &gt;At the time, the team was looking to do some
              optimizations, but I'm not<br>
              &gt;sure this has happened.<br>
              &gt;<br>
              &gt;What can we do to try to improve performance?<br>
              &gt;<br>
              &gt;Thank you.<br>
              &gt;<br>
              &gt;<br>
              &gt;<br>
              &gt;Some setup values follow.<br>
              &gt;<br>
              &gt;xfs_info /mnt/SNIP_block1<br>
              &gt;meta-data=/dev/sdc               isize=512   
              agcount=103,<br>
              &gt;agsize=26214400<br>
              &gt;blks<br>
              &gt;         =                       sectsz=512   attr=2,
              projid32bit=1<br>
              &gt;      =                       crc=1        finobt=1,
              sparse=0, rmapbt=0<br>
              &gt;         =                       reflink=0<br>
              &gt;data     =                       bsize=4096 
               blocks=2684354560,<br>
              &gt;imaxpct=25<br>
              &gt;         =                       sunit=0      swidth=0
              blks<br>
              &gt;naming   =version 2              bsize=4096 
               ascii-ci=0, ftype=1<br>
              &gt;log      =internal log           bsize=4096 
               blocks=51200, version=2<br>
              &gt;        =                       sectsz=512   sunit=0
              blks, lazy-count=1<br>
              &gt;realtime =none                   extsz=4096 
               blocks=0, rtextents=0<br>
              &gt;<br>
              &gt;Volume Name: SNIP_data1<br>
              &gt;Type: Replicate<br>
              &gt;Volume ID: SNIP<br>
              &gt;Status: Started<br>
              &gt;Snapshot Count: 0<br>
              &gt;Number of Bricks: 1 x 4 = 4<br>
              &gt;Transport-type: tcp<br>
              &gt;Bricks:<br>
              &gt;Brick1: nexus2:/mnt/SNIP_block1/SNIP_data1<br>
              &gt;Brick2: forge:/mnt/SNIP_block1/SNIP_data1<br>
              &gt;Brick3: hive:/mnt/SNIP_block1/SNIP_data1<br>
              &gt;Brick4: citadel:/mnt/SNIP_block1/SNIP_data1<br>
              &gt;Options Reconfigured:<br>
              &gt;cluster.quorum-count: 1<br>
              &gt;cluster.quorum-type: fixed<br>
              &gt;network.ping-timeout: 5<br>
              &gt;network.remote-dio: enable<br>
              &gt;performance.rda-cache-limit: 256MB<br>
              &gt;performance.readdir-ahead: on<br>
              &gt;performance.parallel-readdir: on<br>
              &gt;network.inode-lru-limit: 500000<br>
              &gt;performance.md-cache-timeout: 600<br>
              &gt;performance.cache-invalidation: on<br>
              &gt;performance.stat-prefetch: on<br>
              &gt;features.cache-invalidation-timeout: 600<br>
              &gt;features.cache-invalidation: on<br>
              &gt;cluster.readdir-optimize: on<br>
              &gt;performance.io-thread-count: 32<br>
              &gt;server.event-threads: 4<br>
              &gt;client.event-threads: 4<br>
              &gt;performance.read-ahead: off<br>
              &gt;cluster.lookup-optimize: on<br>
              &gt;performance.cache-size: 1GB<br>
              &gt;cluster.self-heal-daemon: enable<br>
              &gt;transport.address-family: inet<br>
              &gt;nfs.disable: on<br>
              &gt;performance.client-io-threads: on<br>
              &gt;cluster.granular-entry-heal: enable<br>
              &gt;cluster.data-self-heal-algorithm: full<br>
              &gt;<br>
              &gt;Sincerely,<br>
              &gt;Artem<br>
              &gt;<br>
              &gt;--<br>
              &gt;Founder, Android Police &lt;<a
                href="http://www.androidpolice.com" rel="noreferrer
                noreferrer" target="_blank" moz-do-not-send="true">http://www.androidpolice.com</a>&gt;,
              APK Mirror<br>
              &gt;&lt;<a href="http://www.apkmirror.com/"
                rel="noreferrer noreferrer" target="_blank"
                moz-do-not-send="true">http://www.apkmirror.com/</a>&gt;,
              Illogical Robot LLC<br>
              &gt;<a href="http://beerpla.net" rel="noreferrer
                noreferrer" target="_blank" moz-do-not-send="true">beerpla.net</a>
              | @ArtemR &lt;<a href="http://twitter.com/ArtemR"
                rel="noreferrer noreferrer" target="_blank"
                moz-do-not-send="true">http://twitter.com/ArtemR</a>&gt;<br>
              <br>
              Hi Artem,<br>
              <br>
              Have you checked the same on brick level ? How big is the
              difference ?<br>
              <br>
              Best Regards,<br>
              Strahil Nikolov<br>
            </blockquote>
          </div>
        </div>
      </div>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <pre class="moz-quote-pre" wrap="">________



Community Meeting Calendar:

Schedule -
Every 2nd and 4th Tuesday at 14:30 IST / 09:00 UTC
Bridge: <a class="moz-txt-link-freetext" href="https://bluejeans.com/441850968">https://bluejeans.com/441850968</a>

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>
  </body>
</html>