[Swift-devel] Re: Swift on Eureka

[Tim Armstrong]

I ran into this bug too but now have coasters working on Eureka.  My
solution was to generate a script to execute to cobalt with the argument
embedded:

cat > $HOME/batch.sub <<END
#!/bin/sh
export WORKER_LOGGING_LEVEL=$workerLogging
echo '***' COBALT_NODEFILE file: \$COBALT_NODEFILE CONTACT:$CONTACT
cat \$COBALT_NODEFILE
echo '***' unique nodes are:
sort < \$COBALT_NODEFILE|uniq
for h in \$(sort < \$COBALT_NODEFILE|uniq); do
  ssh \$h "echo Swift R startup running on host; hostname; cd /;
/usr/bin/perl $SWIFTBIN/worker.pl $CONTACT SwiftR-\$h $HOME/.globus/coasters
$IDLETIMEOUT" &
done
wait

END
chmod +x $HOME/batch.sub

Its kind of hacky but it does work.  Also, inconsistently with cobalt's
behaviour on BG/P, as far as I can tell on Eureka (all clusters?) it behaves
a lot like PBS in that it only runs the submitted executable on one cluster
node, and provides a list of other cluster nodes through an environment
variable.

- Tim

On Sat, Jan 8, 2011 at 4:40 PM, Michael Wilde <wilde at mcs.anl.gov> wrote:

> Thanks, Justin.  cc'ing back to the list, Rob, and Sheri.
>
> Sheri, maybe you can run on PADS or Fusion till this is fixed?
>
> - Mike
>
> ----- Original Message -----
> > Hello
> > Right, Swift does not currently run on Eureka due to the following
> > bug in Cobalt:
> >
> > http://trac.mcs.anl.gov/projects/cobalt/ticket/462
> >
> > I got about half of a work-around for this done...
> >
> > Justin
> >
> > On Fri, 7 Jan 2011, Michael Wilde wrote:
> >
> > > Hi Rob and Sheri,
> > >
> > > I don't know the status of Swift on Eureka, but Im eager to see it
> > > running there, so we'll make sure it works.
> > >
> > > A long while back I tried Swift there, and at the time we had a
> > > minor
> > > bug in the Cobalt provider. Justin may have fixed that recently on
> > > the
> > > BG/P's. So Im hoping it either works or has only some
> > > readily-fixable
> > > issues in the way.
> > >
> > > We'll try it and get back to you.
> > >
> > > In the mean time, Sheri, you might want to try a simple hello-world
> > > test
> > > on Eureka, and see if you can progress to replicating what John
> > > Dennis
> > > had done so far.
> > >
> > > Its best to send any errors you get to the swift-user list (which
> > > you
> > > should join) so that everyone on the Swift team is aware f any
> > > issues
> > > you encounter and can offer help.
> > >
> > > You should meet with Justin at Argonne (3rd floor, 240) who can
> > > serve as
> > > your Swift mentor.
> > >
> > > Sarah, David - lets add Eureka to the test matrix for release 0.92.
> > > Cobalt is very very close to PBS's interface, but there is a
> > > separate
> > > Swift execution provider that handles the differences.
> > >
> > > Regards,
> > >
> > > Mike
> > >
> > >
> > > ----- Original Message -----
> > >> Hi Mike,
> > >>
> > >> Sheri is going to take over some of the development work John
> > >> Dennis
> > >> was
> > >> doing on using swift with the AMWG diag package.
> > >>
> > >> Our platform is Eureka. Is there a development version of Swift
> > >> installed there?
> > >>
> > >> Rob
> > >
> > >
> >
> > --
> > Justin M Wozniak
>
> --
> Michael Wilde
> Computation Institute, University of Chicago
> Mathematics and Computer Science Division
> Argonne National Laboratory
>
> _______________________________________________
> Swift-devel mailing list
> Swift-devel at ci.uchicago.edu
> http://mail.ci.uchicago.edu/mailman/listinfo/swift-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/swift-devel/attachments/20110216/39b300e7/attachment.html>