Re: [ptp-dev] PTP Compiling causes other nodes on cluster to be unavaila

[Greg Watson <g.watson@xxxxxxxxxxxx>]

Mark,

In order for Open MPI to run processes on your other nodes, you have  
to add their host names to OMPI_INSTALL_DIR/etc/openmpi-default- 
hostfile. Currently PTP does not support launching to an arbitrary  
set of nodes (this will be added in 2.0).

There is nothing in the PTP install process that will change an  
existing Open MPI installation.

Regards,

Greg

On Nov 9, 2007, at 1:44 AM, Mark Capogreco wrote:

> Hi All,
>
>
>
> I have spent some time with PTP over the last 3 weeks as looking to  
> use it as a framework for some research work.
>
>
>
> By following the install process using CDT and PTP I got everything  
> up and running except that in PTP Runtime it only shows me the node  
> on which I run eclipse and only allows me to run my processes on  
> this node.  I am working on an 8 node linux x86_64 cluster.
>
>
>
> The problem is that after the install process I can’t run my  
> openMPI 1.2.4 test program in a bash shell using  any other nodes,  
> only the node I am on.  For eg: mpirun –np 8 –host  
> currentNode,othernode1,othernode2 test_program causes the shell to  
> freeze.  If I have eclipse running the error message appears in the  
> eclipse log, saying that the othernode1 and 2 basically don’t exist.
>
>
>
>  If I ssh to another node where I haven’t  done the install process  
> and run my test program it runs fine across several nodes.
>
>
>
> I was wondering if you can tell me why the install process is  
> corrupting my existing openMPI setup, not allowing me to run my  
> test program across other nodes and why I am only seeing 1 machine  
> in my PTP runtime.
>
>
>
> Any help greatly appreciated as really would like to get to the  
> bottom of this.
>
>
>
> Thanks
>
>
>
> Mark
> >