[Swift-devel] coastersPerNode not recognized by GT2 GRAM

[Mihael Hategan]

On Wed, 2008-07-30 at 09:42 -0500, Michael Wilde wrote:
> I need to set this aside for now, but would appreciate any help in 
> debugging it.
> 
> My sites.xml file is:
> 
> <config>
> <pool handle="abe" >
>    <execution provider="coaster" url="grid-abe.ncsa.teragrid.org" 
> jobManager="gt2:gt2:pbs" />
>    <gridftp url="gsiftp://gridftp-abe.ncsa.teragrid.org"/>
>    <workdirectory>/u/ac/wilde/swiftwork</workdirectory>
>    <profile namespace="globus" key="coastersPerNode">8</profile>
> </pool>
> </config>
> 
> The logs are on CI net at /home/wilde/coast/run17.
> 
> I dont see how this works. Is the code picking up the parameter from the 
>   globus RSL and then passing it to bootstrap.sh to in turn pass it to 
> the coaster server? It needs to be stripped off the globus profile 
> before the GT2 job that launches bootstrap.sh is run, right?  Else 
> Globus will complain that its not valid RSL?

Right. Actually the coaster code should explicitly avoid passing that
attribute to gt2. So I consider this a bug.

> 
> I see the one test case for this in tests/sites/coaster is for 
> localhost. Was it tested on gt2:gt2:pbs?
> 
> - Mike
> 
> 
> 
> On 7/30/08 6:42 AM, Michael Wilde wrote:
> > When I use coastersPerNode on Abe I get an error:
> > 
> > 2008-07-30 01:06:43,498-0500 DEBUG vdl:execute2 APPLICATION_EXCEPTION 
> > jobid=echo-9g3mu8xi - Application exception: Cannot submit job
> > org.globus.cog.abstraction.impl.common.task.TaskSubmissionException: 
> > Cannot submit job
> > at 
> > org.globus.cog.abstraction.impl.execution.gt2.JobSubmissionTaskHandler.submitSingleJob(JobSubmissionTaskHandler.java:162) 
> > 
> > ...
> > Caused by: org.globus.gram.GramException: Parameter not supported
> > 
> > Do the globus namespace options get put into RSL in GRAM2, and do these 
> > need to be valid Globus options?
> > 
> > I assume this option is in the globus namespace because that was a 
> > convenient way of adding per-site options?  Or was this mean to be in 
> > the karajan namespace?
> > 
> > I'll dig deeper, but wonder if this has been tested on TeraGrid sites.
> > 
> > - Mike
> > 
> > 
> > On 7/27/08 2:58 PM, Ben Clifford wrote:
> >  > cog svn r2094 introduces a profile property coastersPerNode which allows
> >  > you to spawn multiple coaster workers on a node. this should allow 
> > you to
> >  > take advantage of sites which have multicore CPUs but allocate the whole
> >  > node, rather than an individual core, when a job is submitted.
> >  >
> >  > When using coasters, add this to the site definition:
> >  >
> >  >     <profile namespace="globus" key="coastersPerNode">5</profile>
> >  >
> >  >
> >  > to get eg 5 workers on each node.
> >  >
> > _______________________________________________
> > Swift-devel mailing list
> > Swift-devel at ci.uchicago.edu
> > http://mail.ci.uchicago.edu/mailman/listinfo/swift-devel
> _______________________________________________
> Swift-devel mailing list
> Swift-devel at ci.uchicago.edu
> http://mail.ci.uchicago.edu/mailman/listinfo/swift-devel