[mpich-discuss] Problems with running make

[Gustavo Correa]

Hi Reuti

Thank you for the helpful tips about the ssh setup to enable multi-user MPI.
We have clusters here, which are mostly homogeneous environments,
so in general the user of an MPI program is the same across all computers.
However, in situations where the environment is less homogeneous (say 
groups of PCs in a subnet), the same user with different usernames/IDs
is not uncommon, hence your setup would fit nicely.

I also read, and used, your host-based ssh authentication howto in the past,
and it is also a very helpful resource.

Best,
Gus Correa

On Mar 2, 2014, at 5:41 PM, Reuti wrote:

> Hi,
> 
> Am 02.03.2014 um 22:46 schrieb Gustavo Correa:
> 
>> Hi Pavan
>> 
>> Many thanks for this clarification.  
>> I stand corrected.
>> 
>> Actually I guessed you MPICH developers might have already set 
>> something up to allow this kind of "collaborative" or "dialogue" MPI style
>> with different users names/IDs running MPI processes across different computers, 
>> for specific uses.
>> 
>> Not trying to hijack Ron's thread (he has more important hurdles to get behind him),
>> I wonder how mpiexec (Hydra), which is launched on one computer under
>> one user name, will trigger additional MPI processes on the other computers
>> under different user names (and how ssh may have to be setup to allow this to happen).
> 
> It's necessary to put your public key into the authorized_keys file under the other accounts on the slave machines. I have different accounts at several remote machines and to ease the access, I also put the target account in my ~/.ssh/config like:
> 
> $ cat ~/.ssh/config
> Host slave1
>   User foo
> 
> Host slave2
>   User baz
> 
> Then it will use the specified account automatically. AFAICS it also supports running MPICH applications with different account on a cluster this way without any support by MPICH.
> 
> ==
> 
> You can also use hostbased authentication, for which I wrote a Howto some time ago:
> 
> https://arc.liv.ac.uk/SGE/howto/hostbased-ssh.html
> 
> In case of a different account it must be granted by a setting in ~/.rhosts resp. ~/.shosts in addition.
> 
> 
> -- Reuti
> 
> 
>> Cheers,
>> Gus
>> 
>> On Mar 2, 2014, at 12:39 PM, Balaji, Pavan wrote:
>> 
>>> Gus,
>>> 
>>> On Mar 2, 2014, at 10:43 AM, Gustavo Correa <gus at ldeo.columbia.edu> wrote:
>>>> Rajeev may correct me if I am wrong, 
>>>> but I think you *must* run MPI as the same user (name and UID) everywhere.
>>> 
>>> Not necessarily.
>>> 
>>> http://wiki.mpich.org/mpich/index.php/Using_the_Hydra_Process_Manager#Using_Hydra_on_Machines_with_Different_User_Names
>>> 
>>> But using the same username is definitely easier and safer.
>>> 
>>> — Pavan
>>> 
>>> _______________________________________________
>>> discuss mailing list     discuss at mpich.org
>>> To manage subscription options or unsubscribe:
>>> https://lists.mpich.org/mailman/listinfo/discuss
>> 
>> _______________________________________________
>> discuss mailing list     discuss at mpich.org
>> To manage subscription options or unsubscribe:
>> https://lists.mpich.org/mailman/listinfo/discuss
>> 
> 
> _______________________________________________
> discuss mailing list     discuss at mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss