Am 01.01.2014 um 20:25 schrieb Jeff Hammond:

Isn't the process manager infrastructure hetero-safe? At the very least, ssh-ing "uname -a" around the ring of procs can identify the problem on O(1) cost.

As there are clusters without `ssh` availability (i.e. MPICH is running tightly integrated in a queuing system), I would put this in the initial startup call to the slave daemons: `hydra_pmi_proxy` could get an additional parameter about the architecture of the machine where `mpiexec` was issued. It could then return instantly with an error code for something like "refuse to participate". -- Reuti

Jeff

Sent from my iPhone

...

On Jan 1, 2014, at 11:33 AM, Pavan Balaji <[email protected]> wrote:

If we detect what the local architecture is, how will we communicate it to other processes without having some infrastructure that is hetergeneous-architecture safe?

— Pavan

...

On Jan 1, 2014, at 9:50 AM, Jeff Hammond <[email protected]> wrote:

One wonders if MPICH could verify the system was sufficiently homogeneous to function properly during initialization rather than defer detection to an error during communication calls.

Jeff

Sent from my iPhone

...

On Jan 1, 2014, at 8:07 AM, Matthias Neuer <[email protected]> wrote:

Am 31.12.2013 18:44, schrieb Pavan Balaji:

...

On Dec 31, 2013, at 10:11 AM, Matthias Neuer <[email protected]> wrote:

...

...

On 12/31/2013 04:34 PM, Pavan Balaji wrote:

Are matze-debian and notebook of the same architecture (x86_64 and running the same OS/configuration)?

No, notebook is x86_32 on debian stable and matze-debian is x86_64 on debian testing, but I compiled the program for each system separately.

Such heterogeneous configurations are not supported in mpich.

Ok, thats the problem then. I didn't find that information in the FAQ or in the installation manual.

Thanks for your help. Matthias _______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

---

discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

-- Pavan Balaji http://www.mcs.anl.gov/~balaji

_______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

---

discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss