[MPICH] Problems installing MPICH2 on single machine

[James Frye]

Rajeev,

Yes and no:  When I have a DHCP network connection, I can do an ssh to any 
outside machine on which I have an account.  I can also do an ssh to 
myself with "ssh localhost".  What I can't do is an "ssh a.me.org" - that 
being the private host & domain name I use with my home network (which is 
never connected to the outside world).

That's why I think there must be some iteraction between MPI and the (to 
me mysterious) DHCP network setup.  Even though I tell MPI explicitly (via 
the mpd.hosts file) to use "localhost" as the one and only machine to run 
on, it finds "a.me.org" somewhere, tries to ssh to that instead, and 
hangs.

So there seem to be two possible options: either fix whatever's amiss with 
the mpd, or ask some guru how to change the network setup so it'll work. 
This mailing list is my best shot at either :-)

I'd imagine there must be quite a few people with similar setups, doing 
development on a laptop or single-processor desktop machine, so maybe the 
situation should be addressed in the install instructions.

Thanks,
James


On Wed, 13 Jul 2005, Rajeev Thakur wrote:

>>    ** ssh timed out to a.me.org
>>    ** ssh failed to a.me.org
>
> These lines from mpdcheck suggest that ssh is not working right on your
> machine.
>
> Rajeev
>
>
>> -----Original Message-----
>> From: owner-mpich-discuss at mcs.anl.gov
>> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of James Frye
>> Sent: Monday, July 11, 2005 8:25 AM
>> To: mpich-discuss at mcs.anl.gov
>> Subject: [MPICH] Problems installing MPICH2 on single machine
>>
>> Hi,
>>
>> I'm having problems setting up MPICH2 (and MPICH) to run on a single
>> machine, and would be grateful for any help.
>>
>> The machine is a Dell laptop with P3 processor, running
>> Linux.  Output of
>> uname -a is:
>>
>> Linux a.me.org 2.6.5-1.358 #1 Sat May 8 09:04:50 EDT 2004
>> i686 i686 i386 GNU/Linux
>>
>> The machine is used for development.  It needs to run MPI
>> only on itself,
>> but sometimes with multiple "processors" (e.g.  "mpirun -np 4
>> program") to
>> test code that will eventually run on a parallel machine.
>> I'm often not
>> connected to any network, so I've arbitrarily set HOSTNAME to
>> "a.me.org".
>> When I am connected, it's often via a DHCP connection, so
>> AFAIK (I'm not a
>> network guru) there's no way to give it a real, fixed name.
>>
>> I've configured & installed per instructions, in directory
>> /opt/mpich2/gcc.  Paths & environment are set by sourcing the script
>>
>>    #! /bin/csh
>>
>>    echo 'Setting MPICH2/gcc paths'
>>    setenv MPICH2 /opt/mpich2/gcc
>>    setenv PATH $PATH\:${MPICH2}/bin
>>    setenv MANPATH $MANPATH\:${MPICH2}/man
>>
>> "mpd &" starts, "mpdtrace -l" responds with "a.me.org_32811".
>>
>> "mpiexec -n 1 /bin/hostname" will eventually time out with
>> messages like
>>
>>    a.me.org_mpdman_1: conn error in connect_lhs: Connection timed out
>>    a.me.org_mpdman_0: mpd_uncaught_except_tb handling:
>>      socket_error: (110, 'Connection timed out')
>>    ...
>>
>> "mpdcheck" and "mpdcheck -f mpd.hosts" give no output.
>> "mpdcheck -f mpd.hosts -ssh"
>> gives
>>
>>    ** ssh timed out to a.me.org
>>    ** ssh failed to a.me.org
>>    ** here is the output:
>>
>> (but there is no output.)
>>
>> If instead I use "localhost" in the mpd.hosts file, I get
>>
>>    ** Timed out waiting for client on localhost to produce output.
>>    client on localhost failed to access the server.
>>    here is the output:
>>
>> and again, no output.
>>
>> ---
>>
>> I get similar problems with MPICH (the most recent version,
>> downloaded &
>> installed yesterday).  It tries to use rsh instead of ssh,
>> and can only
>> run with "-np 1".
>>
>> The problem would seem to be related to the DHCP connection.  I've run
>> MPICH for a couple of years on my network at home without
>> problems, but
>> they're all connected to each other and not the outside world.
>> Unfortunately they're about 6000 miles away at the moment,
>> and I need to
>> get some work done...
>>
>> Thanks,
>> James
>>
>>
>