<html>
  <head>
    <meta content="text/html; charset=utf-8" http-equiv="Content-Type">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    Tim, Justin, have you done the static build on Titan, and if so, has
    it resolved the problem?<br>
    <br>
    - Mike<br>
    <br>
    <div class="moz-cite-prefix">On 10/15/14 5:22 PM, Justin M Wozniak
      wrote:<br>
    </div>
    <blockquote
      cite="mid:65cdff0a-1ef6-4d8b-a579-a1fbd762dc60@email.android.com"
      type="cite">
      <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
      <br>
      Bisection showed that loading the Turbine so from plain Tcl breaks
      the OS. <br>
      <br>
      I had OLCF reset my shell to a blank bash so I can make pure use
      of modules. I added support for this in the configuration.<br>
      <br>
      I have been in touch with Tim today about doing a static build on
      the Cray.<br>
      <br>
      <br>
      <div class="gmail_quote">On October 15, 2014 4:31:50 PM CDT,
        Michael Wilde <a class="moz-txt-link-rfc2396E" href="mailto:wilde@anl.gov"><wilde@anl.gov></a> wrote:
        <blockquote class="gmail_quote" style="margin: 0pt 0pt 0pt
          0.8ex; border-left: 1px solid rgb(204, 204, 204);
          padding-left: 1ex;">
          <div class="moz-forward-container">Status as of Monday (but I
            know Justin did a few commits since...)<br>
            <br>
            -------- Forwarded Message --------
            <table class="moz-email-headers-table" border="0"
              cellpadding="0" cellspacing="0">
              <tbody>
                <tr>
                  <th align="RIGHT" nowrap="nowrap" valign="BASELINE">Subject:

                  </th>
                  <td>Re: Latest on the Swift/T Titan problem?</td>
                </tr>
                <tr>
                  <th align="RIGHT" nowrap="nowrap" valign="BASELINE">Date:
                  </th>
                  <td>Mon, 13 Oct 2014 16:11:31 -0500</td>
                </tr>
                <tr>
                  <th align="RIGHT" nowrap="nowrap" valign="BASELINE">From:
                  </th>
                  <td>Justin M Wozniak <a moz-do-not-send="true"
                      class="moz-txt-link-rfc2396E"
                      href="mailto:wozniak@mcs.anl.gov"><wozniak@mcs.anl.gov></a></td>
                </tr>
                <tr>
                  <th align="RIGHT" nowrap="nowrap" valign="BASELINE">To:
                  </th>
                  <td>Michael Wilde <a moz-do-not-send="true"
                      class="moz-txt-link-rfc2396E"
                      href="mailto:wilde@anl.gov"><wilde@anl.gov></a></td>
                </tr>
              </tbody>
            </table>
            <br>
            <br>
            <pre>Pure shell: works
Pure Tcl: works
ADLB batcher: works
<------->
Swift/T: fails

Something in Turbine breaks the OS.  I have been bisecting over a 
stripped-down Turbine to identify exactly what this is.  If I strip down 
the Turbine link step, I can get it to work.  So I think it is something 
in our use of Cray modules.  (I try to not use modules/softenv but it is 
affecting the link.)

This may fix the Swift/NAMD issue, which would be a complementary 
positive outcome.

On 10/13/2014 12:10 PM, Michael Wilde wrote:
> One avenue I was wondering about for debugging:  What PrgEnv is the 
> Swift/T on Titan built for?
> Im assuming gnu, but if so, how does that interact with the 
> PrgEnv-intel needed for the DISCUS app?
>
> I would think to handle such cases we need to run Turbine under the 
> PrgEnv it needs, and then run any app( ) calls that needs a different 
> PrgEnv under a wrapper shell script that provides it, just for that 
> process.
>
> Which may in turn raise the issue of whether "module load 
> PrgEnv-intel" for example is scalable to hundredes to tens of 
> thousands of concurrent invocations.
>
> - Mike
>
> On 10/13/14 11:38 AM, Justin M Wozniak wrote:
>>
>> Yes, I am testing launching tasks with a plain shell script and a 
>> plain Tcl script.  I will then move to the ADLB batcher program.
>>
>> On 10/13/2014 11:34 AM, Michael Wilde wrote:
>>> Hi Justin,
>>>
>>> I did not get to spend any time on this over the weekend.
>>>
>>> What is your latest assessment of the issue?
>>>
>>> I did some runs Fri night that were disturbing: two runs of 256 
>>> sleeps on 128 cores, each yielding a different number of output 
>>> files (but missing ~ 5 in one case and ~ 10 in the other).
>>>
>>> I was going to do a test of a simple MPI app forking tasks with 
>>> system( ), to see if same failure pattern is seen or not, as #CPUs 
>>> increases from 128 to 1024.
>>>
...

-- 
Justin M Wozniak

</pre>
            <br>
          </div>
          <div class="moz-cite-prefix">On 10/15/14 4:29 PM, Michael
            Wilde wrote:<br>
          </div>
          <blockquote cite="mid:543EE736.5020301@anl.gov" type="cite">
            Justin, can you report on your latest findings on this?<br>
            <br>
            (I will fwd Justin's latest notes from Monday...)<br>
            <br>
            - Mike<br>
            <br>
            <div class="moz-cite-prefix">On 10/15/14 4:27 PM, Tim
              Armstrong wrote:<br>
            </div>
            <blockquote
cite="mid:CAC0jiV4q7HbZfSHf0_-UasZFhTLeWHWeOVoaY=d04GMs4Ss48Q@mail.gmail.com"
              type="cite">
              <div dir="ltr">Did you reach a resolution on this?  I'm
                back in town now and could look at adding the ability to
                retrieve the exit code.<br>
                <br>
                - Tim<br>
              </div>
              <div class="gmail_extra"><br>
                <div class="gmail_quote">On Fri, Oct 3, 2014 at 12:07
                  AM, Michael Wilde <span dir="ltr"><<a
                      moz-do-not-send="true" href="mailto:wilde@anl.gov"
                      target="_blank">wilde@anl.gov</a>></span>
                  wrote:<br>
                  <blockquote class="gmail_quote" style="margin:0 0 0
                    .8ex;border-left:1px #ccc solid;padding-left:1ex">
                    <div bgcolor="#FFFFFF" text="#000000"> Also:
                      identical code seems to run fine at this setting:<br>
                        PPN=16 turbine-cray-run.zsh -s titan.settings -n
                      64 $tcl  # (4 nodes)<br>
                      but fails at this setting:<br>
                        PPN=16 turbine-cray-run.zsh -s titan.settings -n
                      128 $tcl # (8 nodes)<br>
                      <br>
                      Changing the app that calls DISCUS directly back
                      to a shell wrapper around DISCUS shows that when
                      the job  fails the app (shell wrapper) is indeed
                      executed, and logs a message to the app stdout
                      file which I see. But I dont see a message after
                      DISCUS.<br>
                      <br>
                      To test if perhaps many concurrent DISCUS
                      invocations (127) are causing a problem which is
                      fatal to the whole job, I added a random sleep
                      before DISCUS.  This had no effect on the problem.<br>
                      <br>
                      Next I eliminated the actual DISCUS call, and just
                      did a random sleep from 0 to 9 seconds. Here I
                      discovered that all the jobs with sleep > 0
                      failed in the same manner as with discus, while
                      the sleep 0 cases all printed all their echos and
                      exited.<br>
                      <br>
                      So there is something more deeply wrong going on,
                      having nothing to do with the DISCUS app. I'll try
                      to narrow this down, and see if perhaps its
                      happening on other non-Cray systems like Midway.<br>
                      <br>
                      - Mike
                      <div>
                        <div class="h5"><br>
                          <br>
                          <div>On 10/2/14 10:52 PM, Michael Wilde wrote:<br>
                          </div>
                          <blockquote type="cite"> Thanks, Tim. But,
                            stderr and stdout were both redirected to
                            files, and all .err and .out files are
                            empty.<br>
                            <br>
                            I'll double-check that these redirects are
                            working, but in prior runs, I did indeed get
                            app error and output in those files.<br>
                            <br>
                            It looks more to me like the app is failing
                            to launch.<br>
                            <br>
                            - Mike<br>
                            <br>
                            <div>On 10/2/14 10:22 PM, Tim Armstrong
                              wrote:<br>
                            </div>
                            <blockquote type="cite">
                              <div dir="ltr">
                                <div>The app stderr/stdout should go to
                                  Swift/T stderr/stdout unless
                                  redirected.  The problem is most
                                  likely discus returning a non-zero
                                  error code.<br>
                                  <br>
                                </div>
                                - Tim<br>
                              </div>
                              <div class="gmail_extra"><br>
                                <div class="gmail_quote">On Thu, Oct 2,
                                  2014 at 8:42 PM, Michael Wilde <span
                                    dir="ltr"><<a
                                      moz-do-not-send="true"
                                      href="mailto:wilde@anl.gov"
                                      target="_blank">wilde@anl.gov</a>></span>
                                  wrote:<br>
                                  <blockquote class="gmail_quote"
                                    style="margin:0 0 0
                                    .8ex;border-left:1px #ccc
                                    solid;padding-left:1ex"><br>
                                    Im getting a failure from a Swift/T
                                    run on Titan that I cant diagnose.<br>
                                    <br>
                                    The script is running 16K DISCUS
                                    simulations, about 30 secs each, 16
                                    per node, on 8 nodes. 127 workers, 1
                                    Server. It seems to run well at
                                    smaller scale (eg 1K simulations).<br>
                                    <br>
                                    My PBS output file shows that some
                                    of the initial round of simulations
                                    fail.  It seems that when this
                                    happens, the Swift script exits
                                    within the first round of
                                    simulations, and none seem to even
                                    start.<br>
                                    <br>
                                    Some of the logs are below.  I'll
                                    continue to experiment and add more
                                    logging to try to isolate the cause.<br>
                                    <br>
                                    I was initially suspicious of an OOM
                                    problem, but I dont see any sign of
                                    that.<br>
                                    <br>
                                    In the line below "Swift: external
                                    command failed:
                                    /lustre/atlas/proj-shared/mat049/scu/build/DiffuseCode/titan/discus/prog/discus





                                    --macro ../stacking-faults3.mac
                                    0.004 0.264 1 0.75 ../Si.cll
                                    0.004_0.264.inte 0.004_0.264.stru" -
                                    is there a way to tell what Swift
                                    things the exit status of these
                                    failing apps are?  What is causing
                                    the whole run to abort?<br>
                                    <br>
                                    It almost looks like one app, or a
                                    small number of them, are generating
                                    a SIGABRT, but thats not clear, and
                                    I have no stdout/err info from the
                                    app processes.<br>
                                    <br>
                                    I'll look for some more detailed
                                    debug options, and/or add some
                                    wrappers around the apps.<br>
                                    <br>
                                    Thanks,<br>
                                    <br>
                                    - Mike<br>
                                    <br>
                                    ---<br>
                                    <br>
                                    T$ cat output.txt.2115044.out<br>
                                    <br>
                                    Swift: external command failed:
                                    /lustre/atlas/proj-shared/mat049/scu/build/DiffuseCode/titan/discus/prog/discus





                                    --macro ../stacking-faults3.mac
                                    0.004 0.264 1 0.75 ../Si.cll
                                    0.004_0.264.inte 0.004_0.264.stru<br>
                                    <br>
                                    Swift: killing MPI job...<br>
                                    ADLB_Abort(1)<br>
                                    MPI_Abort(1)<br>
                                    <br>
                                    Swift: external command failed:
                                    /lustre/atlas/proj-shared/mat049/scu/build/DiffuseCode/titan/discus/prog/discus





                                    --macro ../stacking-faults3.mac
                                    0.004 0.272 1 0.75 ../Si.cll
                                    0.004_0.272.inte 0.004_0.272.stru<br>
                                    <br>
                                    Swift: killing MPI job...<br>
                                    ADLB_Abort(1)<br>
                                    MPI_Abort(1)<br>
                                    <br>
                                    Swift: external command failed:
                                    /lustre/atlas/proj-shared/mat049/scu/build/DiffuseCode/titan/discus/prog/discus





                                    --macro ../stacking-faults3.mac 0.26
                                    0.052000000000000005 1 0.75
                                    ../Si.cll 0.260_0.052.inte
                                    0.260_0.052.stru<br>
                                    <br>
                                    Swift: killing MPI job...<br>
                                    ADLB_Abort(1)<br>
                                    MPI_Abort(1)<br>
                                    <br>
                                    Swift: external command failed:
                                    /lustre/atlas/proj-shared/mat049/scu/build/DiffuseCode/titan/discus/prog/discus





                                    --macro ../stacking-faults3.mac
                                    0.004 0.084 1 0.75 ../Si.cll
                                    0.004_0.084.inte 0.004_0.084.stru<br>
                                    <br>
                                    Swift: killing MPI job...<br>
                                    ADLB_Abort(1)<br>
                                    MPI_Abort(1)<br>
                                    <br>
                                    Swift: external command failed:
                                    /lustre/atlas/proj-shared/mat049/scu/build/DiffuseCode/titan/discus/prog/discus





                                    --macro ../stacking-faults3.mac 0.26
                                    0.024 1 0.75 ../Si.cll
                                    0.260_0.024.inte 0.260_0.024.stru<br>
                                    <br>
                                    Swift: killing MPI job...<br>
                                    ADLB_Abort(1)<br>
                                    MPI_Abort(1)<br>
                                    <br>
                                    Swift: external command failed:
                                    /lustre/atlas/proj-shared/mat049/scu/build/DiffuseCode/titan/discus/prog/discus





                                    --macro ../stacking-faults3.mac
                                    0.004 0.056 1 0.75 ../Si.cll
                                    0.004_0.056.inte 0.004_0.056.stru<br>
                                    <br>
                                    Swift: killing MPI job...<br>
                                    ADLB_Abort(1)<br>
                                    MPI_Abort(1)<br>
                                    <br>
                                    Swift: external command failed:
                                    /lustre/atlas/proj-shared/mat049/scu/build/DiffuseCode/titan/discus/prog/discus





                                    --macro ../stacking-faults3.mac
                                    0.004 0.364 1 0.75 ../Si.cll
                                    0.004_0.364.inte 0.004_0.364.stru<br>
                                    <br>
                                    Swift: killing MPI job...<br>
                                    ADLB_Abort(1)<br>
                                    MPI_Abort(1)<br>
                                    <br>
                                    Swift: external command failed:
                                    /lustre/atlas/proj-shared/mat049/scu/build/DiffuseCode/titan/discus/prog/discus





                                    --macro ../stacking-faults3.mac
                                    0.004 0.07200000000000001 1 0.75
                                    ../Si.cll 0.004_0.072.inte
                                    0.004_0.072.stru<br>
                                    <br>
                                    Swift: killing MPI job...<br>
                                    ADLB_Abort(1)<br>
                                    MPI_Abort(1)<br>
                                    <br>
                                    Swift: external command failed:
                                    /lustre/atlas/proj-shared/mat049/scu/build/DiffuseCode/titan/discus/prog/discus





                                    --macro ../stacking-faults3.mac 0.26
                                    0.044 1 0.75 ../Si.cll
                                    0.260_0.044.inte 0.260_0.044.stru<br>
                                    <br>
                                    Swift: killing MPI job...<br>
                                    ADLB_Abort(1)<br>
                                    MPI_Abort(1)<br>
                                    Application 7706378 exit codes: 134<br>
                                    Application 7706378 exit signals:
                                    Killed<br>
                                    Application 7706378 resources: utime
                                    ~5s, stime ~19s, Rss ~11016,
                                    inblocks ~48986, outblocks ~205<br>
                                    T$<br>
                                    <br>
                                    ---<br>
                                    <br>
                                    T$ cat output.txt<br>
                                    bash_profile: loading modules<br>
                                    DB /ccs/home/wildemj/.modules:
                                    loading modules...<br>
                                    Turbine: turbine-aprun.sh<br>
                                    10/02/2014 09:15PM<br>
                                    <br>
                                    TURBINE_HOME:
                                    /lustre/atlas2/mat049/proj-shared/sfw/compute/turbine<br>
                                    SCRIPT:
/lustre/atlas2/mat049/proj-shared/wildemj/swift-discus/stacking-faults3.tcl<br>
                                    PROCS:        128<br>
                                    NODES:        8<br>
                                    PPN:          16<br>
                                    WALLTIME:     00:15:00<br>
                                    <br>
                                    TURBINE_WORKERS: 127<br>
                                    ADLB_SERVERS:    1<br>
                                    <br>
                                    TCLSH:
                                    /lustre/atlas2/mat049/proj-shared/sfw/tcl-8.6.2/bin/tclsh8.6<br>
                                    <br>
                                    JOB OUTPUT:<br>
                                    <br>
                                    Rank 2 [Thu Oct  2 21:15:32 2014]
                                    [c12-0c1s1n3] application called
                                    MPI_Abort(MPI_COMM_WORLD, 1) -
                                    process 2<br>
                                    Rank 4 [Thu Oct  2 21:15:32 2014]
                                    [c14-0c1s1n3] application called
                                    MPI_Abort(MPI_COMM_WORLD, 1) -
                                    process 4<br>
                                    Rank 6 [Thu Oct  2 21:15:32 2014]
                                    [c14-0c1s1n1] application called
                                    MPI_Abort(MPI_COMM_WORLD, 1) -
                                    process 6<br>
                                    Rank 3 [Thu Oct  2 21:15:32 2014]
                                    [c14-0c1s1n2] application called
                                    MPI_Abort(MPI_COMM_WORLD, 1) -
                                    process 3<br>
                                    Rank 5 [Thu Oct  2 21:15:32 2014]
                                    [c14-0c1s1n0] application called
                                    MPI_Abort(MPI_COMM_WORLD, 1) -
                                    process 5<br>
                                    Rank 1 [Thu Oct  2 21:15:32 2014]
                                    [c12-0c1s1n2] application called
                                    MPI_Abort(MPI_COMM_WORLD, 1) -
                                    process 1<br>
                                    Rank 7 [Thu Oct  2 21:15:32 2014]
                                    [c14-0c1s0n0] application called
                                    MPI_Abort(MPI_COMM_WORLD, 1) -
                                    process 7<br>
                                    _pmiu_daemon(SIGCHLD): [NID 06114]
                                    [c14-0c1s1n0] [Thu Oct  2 21:15:32
                                    2014] PE RANK 5 exit signal Aborted<br>
                                    _pmiu_daemon(SIGCHLD): [NID 06076]
                                    [c14-0c1s1n2] [Thu Oct  2 21:15:32
                                    2014] PE RANK 3 exit signal Aborted<br>
                                    _pmiu_daemon(SIGCHLD): [NID 06077]
                                    [c14-0c1s1n3] [Thu Oct  2 21:15:32
                                    2014] PE RANK 4 exit signal Aborted<br>
                                    _pmiu_daemon(SIGCHLD): [NID 04675]
                                    [c12-0c1s1n3] [Thu Oct  2 21:15:32
                                    2014] PE RANK 2 exit signal Aborted<br>
                                    [NID 04675] 2014-10-02 21:15:32 Apid
                                    7706378: initiated application
                                    termination<br>
                                    _pmiu_daemon(SIGCHLD): [NID 06115]
                                    [c14-0c1s1n1] [Thu Oct  2 21:15:32
                                    2014] PE RANK 6 exit signal Aborted<br>
                                    T$<br>
                                    T$<br>
                                    T$<br>
                                    T$ cat turbine.log<br>
                                    JOB:               2115044<br>
                                    COMMAND:         
                                     stacking-faults3.tcl<br>
                                    HOSTNAME:          <a
                                      moz-do-not-send="true"
                                      href="http://ccs.ornl.gov"
                                      target="_blank">ccs.ornl.gov</a><br>
                                    SUBMITTED:         10/02/2014
                                    09:14PM<br>
                                    PROCS:             128<br>
                                    PPN:               16<br>
                                    NODES:             8<br>
                                    TURBINE_WORKERS:<br>
                                    ADLB_SERVERS:<br>
                                    WALLTIME:          00:15:00<br>
                                    ADLB_EXHAUST_TIME:<br>
                                    T$<span><font color="#888888"><br>
                                        <br>
                                        <br>
                                        <br>
                                        <br>
                                        -- <br>
                                        Michael Wilde<br>
                                        Mathematics and Computer
                                        Science          Computation
                                        Institute<br>
                                        Argonne National Laboratory     
                                                 The University of
                                        Chicago<br>
                                        <br>
_______________________________________________<br>
                                        ExM-user mailing list<br>
                                        <a moz-do-not-send="true"
                                          href="mailto:ExM-user@lists.mcs.anl.gov"
                                          target="_blank">ExM-user@lists.mcs.anl.gov</a><br>
                                        <a moz-do-not-send="true"
                                          href="https://lists.mcs.anl.gov/mailman/listinfo/exm-user"
                                          target="_blank">https://lists.mcs.anl.gov/mailman/listinfo/exm-user</a><br>
                                      </font></span></blockquote>
                                </div>
                                <br>
                              </div>
                              <br>
                              <fieldset></fieldset>
                              <br>
                              <pre>_______________________________________________
ExM-user mailing list
<a moz-do-not-send="true" href="mailto:ExM-user@lists.mcs.anl.gov" target="_blank">ExM-user@lists.mcs.anl.gov</a>
<a moz-do-not-send="true" href="https://lists.mcs.anl.gov/mailman/listinfo/exm-user" target="_blank">https://lists.mcs.anl.gov/mailman/listinfo/exm-user</a>
</pre>
                            </blockquote>
                            <br>
                            <pre cols="72">-- 
Michael Wilde
Mathematics and Computer Science          Computation Institute
Argonne National Laboratory               The University of Chicago
</pre>
                          </blockquote>
                          <br>
                          <pre cols="72">-- 
Michael Wilde
Mathematics and Computer Science          Computation Institute
Argonne National Laboratory               The University of Chicago
</pre>
                        </div>
                      </div>
                    </div>
                    <br>
                    _______________________________________________<br>
                    ExM-user mailing list<br>
                    <a moz-do-not-send="true"
                      href="mailto:ExM-user@lists.mcs.anl.gov">ExM-user@lists.mcs.anl.gov</a><br>
                    <a moz-do-not-send="true"
                      href="https://lists.mcs.anl.gov/mailman/listinfo/exm-user"
                      target="_blank">https://lists.mcs.anl.gov/mailman/listinfo/exm-user</a><br>
                    <br>
                  </blockquote>
                </div>
                <br>
              </div>
            </blockquote>
            <br>
            <pre class="moz-signature" cols="72">-- 
Michael Wilde
Mathematics and Computer Science          Computation Institute
Argonne National Laboratory               The University of Chicago
</pre>
          </blockquote>
          <br>
          <pre class="moz-signature" cols="72">-- 
Michael Wilde
Mathematics and Computer Science          Computation Institute
Argonne National Laboratory               The University of Chicago
</pre>
          <pre class="k9mail"><hr>
ExM-user mailing list
<a class="moz-txt-link-abbreviated" href="mailto:ExM-user@lists.mcs.anl.gov">ExM-user@lists.mcs.anl.gov</a>
<a moz-do-not-send="true" href="https://lists.mcs.anl.gov/mailman/listinfo/exm-user">https://lists.mcs.anl.gov/mailman/listinfo/exm-user</a>
</pre>
        </blockquote>
      </div>
      <br>
      -- <br>
      Justin M Wozniak (via phone)
    </blockquote>
    <br>
    <pre class="moz-signature" cols="72">-- 
Michael Wilde
Mathematics and Computer Science          Computation Institute
Argonne National Laboratory               The University of Chicago
</pre>
  </body>
</html>