<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
  <head>
    <meta content="text/html; charset=ISO-8859-1"
      http-equiv="Content-Type">
  </head>
  <body text="#000000" bgcolor="#ffffff">
    <br>
    <br>
    On 03/01/11 09:26, Charles K Pepperdine wrote:
    <blockquote
      cite="mid:3BA76B61-E7DA-49E1-AEFE-819214B3C729@kodewerk.com"
      type="cite">
      <div>Tenured is 524288K of which 478087K is occupied, greater than
        90% I've just been recently poking about in the source trying to
        sort out how logs are being printed. I've seen that partial
        fragments of the log are printed and that the  " [CMS" fragment
        is printed only after a CMS has been triggered. So, the problem
        is bounded by the log messages (sans a stray pointer bug).</div>
      <div><br>
      </div>
      <div>My question is, shouldn't a 90% occupancy of tenured trigger
        a full GC?</div>
    </blockquote>
    <br>
    The CMS concurrent collections are fast from what I've seen (on the
    order of<br>
    the time between ParNew collections).   The rate at which objects
    are getting<br>
    promoted is also low (maybe 3m per ParNew collection).  CMS thinks
    it<br>
    can wait to start a concurrent collection.  The fact that a
    "promotion failure"<br>
    happened makes it look like fragmentation.<br>
    <br>
    <blockquote
      cite="mid:3BA76B61-E7DA-49E1-AEFE-819214B3C729@kodewerk.com"
      type="cite">
      <div>
        <div><br>
        </div>
        <div>Regards,</div>
        <div>Kirk</div>
        <div><br>
        </div>
        <div><br style="font-size: 15px;">
          <div>
            <div>On Mar 1, 2011, at 5:29 PM, Jon Masamitsu wrote:</div>
            <br class="Apple-interchange-newline">
            <blockquote type="cite">
              <div>For 6692906 to be the problem there needs to be a<br>
                CMS concurrent phase in progress (marking, precleaning
                or<br>
                sweeping) and a minor collection running (with<br>
                UseParNewGC in use).  From the fragment of the gc log<br>
                I could not tell for sure (maybe it was in the ...
                removed)<br>
                but I don't think a concurrent phase was in progress<br>
                so I would say it is not 6692906).  Did you try<br>
                -XX:-UseParNewGC as was suggested?  Your minor<br>
                pauses are not particularly long so maybe you<br>
                could afford to try it.  6692906 will not happen<br>
                without UseParNewGC.   Note you need to turn of<br>
                UseParNewGC as it is the default for CMS.<br>
                <br>
                Look back through the log for any other<br>
                ParNew (promotion failed) and see what happens<br>
                in those cases (in you find one).  2+ hours is too<br>
                long.<br>
                <br>
                The gentleman who would know best about this code<br>
                is out of the office until the end of the week.  I'll
                talk<br>
                to him about this to see if he remembers a recent<br>
                fix that I don't.<br>
                <br>
                <br>
                On 03/01/11 01:38, Bogdan Dimitriu wrote:<br>
                <blockquote type="cite">Hi guys,<br>
                </blockquote>
                <blockquote type="cite"><br>
                </blockquote>
                <blockquote type="cite">We're having a problem with
                  garbage collection as described here:<br>
                </blockquote>
                <blockquote type="cite"><a moz-do-not-send="true"
                    href="http://forums.oracle.com/forums/message.jspa?messageID=9345173">http://forums.oracle.com/forums/message.jspa?messageID=9345173</a>
                  (I<br>
                </blockquote>
                <blockquote type="cite">apologise if posting links is
                  not the right policy, but I prefer not to<br>
                </blockquote>
                <blockquote type="cite">duplicate data).<br>
                </blockquote>
                <blockquote type="cite"><br>
                </blockquote>
                <blockquote type="cite">We are going to try an upgrade
                  to JRE 6u24 soon, but reading the release<br>
                </blockquote>
                <blockquote type="cite">notes for each of the versions
                  since 6u20, I don't have much hope of<br>
                </blockquote>
                <blockquote type="cite">this upgrade fixing the problem.<br>
                </blockquote>
                <blockquote type="cite"><br>
                </blockquote>
                <blockquote type="cite">I have searched a bit on the
                  Java bugs database and I've come across<br>
                </blockquote>
                <blockquote type="cite">something that looks similar to
                  the problem I am experiencing:<br>
                </blockquote>
                <blockquote type="cite"><a moz-do-not-send="true"
                    href="http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6692906">http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6692906</a>,
                  but I'm not<br>
                </blockquote>
                <blockquote type="cite">convinced this is exactly the
                  same issue. This bug it seems will be<br>
                </blockquote>
                <blockquote type="cite">fixed in 6u25 (which I've read
                  will be released late March or early April).<br>
                </blockquote>
                <blockquote type="cite"><br>
                </blockquote>
                <blockquote type="cite">The reason I'm leaning towards
                  thinking this is a JVM bug is the fact<br>
                </blockquote>
                <blockquote type="cite">that the JVM can stay in the
                  hung state (as described on the forum) for<br>
                </blockquote>
                <blockquote type="cite">2+ hours until we kill the
                  process.<br>
                </blockquote>
                <blockquote type="cite"><br>
                </blockquote>
                <blockquote type="cite">I was hoping to get an idea
                  about this from the source :), so any hints<br>
                </blockquote>
                <blockquote type="cite">will be greatly appreciated.<br>
                </blockquote>
                <blockquote type="cite"><br>
                </blockquote>
                <blockquote type="cite">Thanks,<br>
                </blockquote>
                <blockquote type="cite">Bogdan<br>
                </blockquote>
                <blockquote type="cite">_______________________________________________<br>
                </blockquote>
                <blockquote type="cite">hotspot-gc-use mailing list<br>
                </blockquote>
                <blockquote type="cite"><a moz-do-not-send="true"
                    href="mailto:hotspot-gc-use@openjdk.java.net">hotspot-gc-use@openjdk.java.net</a><br>
                </blockquote>
                <blockquote type="cite"><a moz-do-not-send="true"
                    href="http://mail.openjdk.java.net/mailman/listinfo/hotspot-gc-use">http://mail.openjdk.java.net/mailman/listinfo/hotspot-gc-use</a><br>
                </blockquote>
                _______________________________________________<br>
                hotspot-gc-use mailing list<br>
                <a moz-do-not-send="true"
                  href="mailto:hotspot-gc-use@openjdk.java.net">hotspot-gc-use@openjdk.java.net</a><br>
                <a moz-do-not-send="true"
                  href="http://mail.openjdk.java.net/mailman/listinfo/hotspot-gc-use">http://mail.openjdk.java.net/mailman/listinfo/hotspot-gc-use</a><br>
              </div>
            </blockquote>
          </div>
          <br>
        </div>
      </div>
    </blockquote>
  </body>
</html>