<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
  <head>
    <meta content="text/html; charset=UTF-8" http-equiv="Content-Type">
  </head>
  <body bgcolor="#ffffff" text="#000000">
    Thanks a lot Stig... your analysis in that discussion goes way
    beyond mind. Did you ever sort it out?<br>
    <br>
    In the meantime I found an unlikely setting that solved my problem:
    the session_linger. I got it from here [1] and thought it wouldn't
    hurt and it nearly killed me. My initial tests were conducted with a
    value of 150, I had to lower it to 20 to get my test through.<br>
    <br>
    Thanks + best regards<br>
    <br>
    [1]
    <a class="moz-txt-link-freetext" href="https://kristianlyng.wordpress.com/2009/10/19/high-end-varnish-tuning/">https://kristianlyng.wordpress.com/2009/10/19/high-end-varnish-tuning/</a><br>
    <br>
    On 05.01.2011 23:56, Stig Bakken wrote:
    <blockquote
      cite="mid:AANLkTincEpY0fOWmLR-FmdgLWD3dPYPUQRPjAAzPEFTv@mail.gmail.com"
      type="cite">This thread: <a moz-do-not-send="true"
href="http://www.varnish-cache.org/lists/pipermail/varnish-misc/2010-December/005258.html">http://www.varnish-cache.org/lists/pipermail/varnish-misc/2010-December/005258.html</a>
      <div><br>
      </div>
      <div> - Stig<br>
        <br>
        <div class="gmail_quote">On Wed, Jan 5, 2011 at 11:44 PM, George
          Georgovassilis <span dir="ltr"><<a moz-do-not-send="true"
              href="mailto:g.georgovassilis@gmail.com">g.georgovassilis@gmail.com</a>></span>
          wrote:<br>
          <blockquote class="gmail_quote" style="margin: 0pt 0pt 0pt
            0.8ex; border-left: 1px solid rgb(204, 204, 204);
            padding-left: 1ex;">
            <div bgcolor="#ffffff" text="#000000"> Hello Stig,<br>
              <br>
              Thanks for the insight. I'm still on the logs, though not
              sure where to start - it's not like that there are any
              errors in it so I'm not really sure what to look for. Do
              you have a pointer to that discussion you are referring
              to?
              <div>
                <div class="h5"><br>
                  <br>
                  On 05.01.2011 23:41, Stig Bakken wrote:
                  <blockquote type="cite">This seems similar to what
                    I've been seeing, described in an earlier thread
                    from before christmas.  In my case it was not during
                    benchmarking, but when serving production load of
                    around 300 req/s per server.  Modern tcpip stacks on
                    modern hardware should handle this without blinking.
                    <div> <br>
                    </div>
                    <div>Did you have the chance to capture the problem
                      with varnishlog so you can replay/analyze it?</div>
                    <div><br>
                    </div>
                    <div> - Stig<br>
                      <br>
                      <div class="gmail_quote">On Wed, Jan 5, 2011 at
                        11:18 PM, George Georgovassilis <span dir="ltr"><<a
                            moz-do-not-send="true"
                            href="mailto:g.georgovassilis@gmail.com"
                            target="_blank">g.georgovassilis@gmail.com</a>></span>
                        wrote:<br>
                        <blockquote class="gmail_quote" style="margin:
                          0pt 0pt 0pt 0.8ex; border-left: 1px solid
                          rgb(204, 204, 204); padding-left: 1ex;">I
                          removed the varnish instance so that the load
                          generator is directly hitting Tomcat.
                          Naturally, the request rate drops to 70
                          requests/sec with a CPU load of 100%...
                          however connections don't drop anymore, no
                          timeouts occur and the application remains
                          pretty responsive. To recap, these are the
                          possible scenarios:<br>
                          <br>
                          1. The networking layer is overtaxed with the
                          original 300 reqs/sec. I don't believe that,
                          because the load generator doesn't record any
                          dropped connections while a simple browser
                          can't connect.<br>
                          <br>
                          2. Tomcat is overtaxed. That also seems not
                          plausible, since it is not servicing any
                          requests under the load test - all is done by
                          varnish. Even if, as I said when removing
                          varnish from in between, it serves the
                          requests just fine.<br>
                          <br>
                          3. Varnish is overtaxed. Somehow that also
                          doesn't make sense, since it is servicing the
                          load generator just fine... but will refuse to
                          serve browser requests.<br>
                          <br>
                          4. Varnish, when under load, is picky about
                          what connections to serve.<br>
                          <br>
                          I'm stuck :-)
                          <div>
                            <div><br>
                              <br>
                              On 05.01.2011 17:59, Bob Camp wrote:<br>
                              <blockquote class="gmail_quote"
                                style="margin: 0pt 0pt 0pt 0.8ex;
                                border-left: 1px solid rgb(204, 204,
                                204); padding-left: 1ex;"> Hi<br>
                                <br>
                                Running simple load tests both on Apache
                                directly, and on Varnish - both<br>
                                seem to experience "long delays" on a
                                small percentage of the requests. The<br>
                                problem does not appear to happen with
                                low loads. It does come up as CPU<br>
                                usage becomes an issue. It also is hard
                                to make happen with a single stream<br>
                                of requests. It seems to come up much
                                quicker with many requests done in<br>
                                parallel.<br>
                                <br>
                                I've always *assumed* that the poor
                                little TCP/IP hamster simply ran out of<br>
                                breath and started dropping connections.<br>
                                <br>
                                Bob<br>
                                <br>
                                -----Original Message-----<br>
                                From: <a moz-do-not-send="true"
                                  href="mailto:varnish-misc-bounces@varnish-cache.org"
                                  target="_blank">varnish-misc-bounces@varnish-cache.org</a><br>
                                [mailto:<a moz-do-not-send="true"
                                  href="mailto:varnish-misc-bounces@varnish-cache.org"
                                  target="_blank">varnish-misc-bounces@varnish-cache.org</a>]
                                On Behalf Of George<br>
                                Georgovassilis<br>
                                Sent: Wednesday, January 05, 2011 11:18
                                AM<br>
                                To: <a moz-do-not-send="true"
                                  href="mailto:varnish-misc@projects.linpro.no"
                                  target="_blank">varnish-misc@projects.linpro.no</a><br>
                                Subject: Re: Connections dropped under
                                load<br>
                                <br>
                                Hello Cosimo,<br>
                                <br>
                                Thank you for the quick reply. After
                                your hint I had the tests run again<br>
                                but couldn't detect that pattern. What
                                susprised me though after looking<br>
                                through the logs is that almost all
                                requests by the load generator<br>
                                complete in a timely manner (<  1
                                sec), but all requests generated by a<br>
                                real browser (IE, FF, Opera) will be
                                served much later or even run into<br>
                                a timeout.<br>
                                <br>
                                On 05.01.2011 16:30, Cosimo Streppone
                                wrote:<br>
                                <blockquote class="gmail_quote"
                                  style="margin: 0pt 0pt 0pt 0.8ex;
                                  border-left: 1px solid rgb(204, 204,
                                  204); padding-left: 1ex;"> On Wed, 05
                                  Jan 2011 16:20:31 +0100, George
                                  Georgovassilis<br>
                                  <<a moz-do-not-send="true"
                                    href="mailto:g.georgovassilis@gmail.com"
                                    target="_blank">g.georgovassilis@gmail.com</a>>

                                   wrote:<br>
                                  <br>
                                  <blockquote class="gmail_quote"
                                    style="margin: 0pt 0pt 0pt 0.8ex;
                                    border-left: 1px solid rgb(204, 204,
                                    204); padding-left: 1ex;"> I'm
                                    having trouble with dropped
                                    connections under a loadtest.<br>
                                    <br>
                                    The problem: As a measure for
                                    response, I am requesting an image
                                    from<br>
                                    the webapp running in Tomcat while
                                    the loadtest is underway. However<br>
                                    that either times out or is
                                    delivered after several seconds.
                                    Varnishlog<br>
                                    will often either not show the
                                    request (RxURL) at all, or show it<br>
                                    several seconds after the browser
                                    dispatched it.<br>
                                  </blockquote>
                                  Hi George,<br>
                                  <br>
                                  if you measure the time you mention as
                                  "several seconds"<br>
                                  and it's either 3 or 9 seconds, I
                                  think what you're seeing<br>
                                  is a client-side TCP retransmit
                                  timeout.<br>
                                  <br>
                                  I experienced that, both under load
                                  testing,<br>
                                  and in real production setups.<br>
                                  <br>
                                </blockquote>
                                <br>
_______________________________________________<br>
                                varnish-misc mailing list<br>
                                <a moz-do-not-send="true"
                                  href="mailto:varnish-misc@varnish-cache.org"
                                  target="_blank">varnish-misc@varnish-cache.org</a><br>
                                <a moz-do-not-send="true"
                                  href="http://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc"
                                  target="_blank">http://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc</a><br>
                                <br>
                              </blockquote>
                              <br>
                              <br>
_______________________________________________<br>
                              varnish-misc mailing list<br>
                              <a moz-do-not-send="true"
                                href="mailto:varnish-misc@varnish-cache.org"
                                target="_blank">varnish-misc@varnish-cache.org</a><br>
                              <a moz-do-not-send="true"
                                href="http://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc"
                                target="_blank">http://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc</a><br>
                            </div>
                          </div>
                        </blockquote>
                      </div>
                      <br>
                      <br clear="all">
                      <br>
                      -- <br>
                      Stig Bakken<br>
                      CTO, Zedge.net - free your phone!<br>
                    </div>
                  </blockquote>
                  <br>
                </div>
              </div>
            </div>
            <br>
            _______________________________________________<br>
            varnish-misc mailing list<br>
            <a moz-do-not-send="true"
              href="mailto:varnish-misc@varnish-cache.org">varnish-misc@varnish-cache.org</a><br>
            <a moz-do-not-send="true"
              href="http://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc"
              target="_blank">http://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc</a><br>
          </blockquote>
        </div>
        <br>
        <br clear="all">
        <br>
        -- <br>
        Stig Bakken<br>
        CTO, Zedge.net - free your phone!<br>
      </div>
    </blockquote>
    <br>
  </body>
</html>