Cc'ing [email protected]. Please keep it cc'ed. On 06/28/2013 12:05 PM, Danilo wrote:

Hi Pavan, so it worked for you with realDim = 16384 and 3 processes.. Can you try with 4 or 8 processes? If so, what could be the problem? Probably the configuration of mpi on the cluster here?

-----Original Message----- From: Pavan Balaji <[email protected]> To: Danilo <[email protected]> Sent: Fri, Jun 28, 2013 6:42 pm Subject: Re: [mpich-discuss] mpi assertion error

Please [email protected] <mailto:[email protected]> cc'ed.

I can run your application fine. I tried 3 processes.

-- Pavan

On 06/28/2013 11:10 AM, Danilo wrote:

...

Hi Pavan,

here it is a simple c file that creates a matrix with 16384 rows, 16384x2 columns and fills it with random float numbers (the columns are double than the rows because they should represent real and imaginary part of a number). Then the master tries to distribute the elements trough scatter/gather. This gives the error as the fft app i posted before..

Thanks for your help

-----Original Message----- From: Pavan Balaji <[email protected] <mailto:[email protected]>> To: discuss <[email protected] <mailto:[email protected]>> Cc: Danilo <[email protected] <mailto:[email protected]>> Sent: Fri, Jun 28, 2013 5:46 pm Subject: Re: [mpich-discuss] mpi assertion error

Danilo,

On 06/28/2013 10:28 AM, Danilo wrote:

...

the program was tested intensively on the previous cluster. The changes made are in scatter/gather (due to sendbuf and recvbuff that has to be differente in this version it seems..). The other main change is due to hydra, because on the previous cluster there wasn't such a process management system. But I'm quite new to programming, so I don't know...

This doesn't look like a Hydra problem. Hydra is just telling you that the application died in an unexpected manner.

The error the assert is showing is the real culprit. Can you try to strip out most of the code and create a simple benchmark that reproduces this error?

-- Pavan

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

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

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