Re: [Myricom help #8330] MPI send/recv bug

Susan Blackford <[email protected]>
Newsgroups gmane.network.myrinet.general
Organization Myricom, Inc.
Message-ID <0112031455120H.00487@idefix>
Hi, David,

On Tuesday 27 November 2001 06:18 pm, David Skinner wrote:
> Hi,
>
> Anyone else experiencing the following bug?
>
> MPI_Send/Recv work from statically allocated memory but are unreliable
> from dynamically allocated memory.
>
> test case:
>
> MPI task 0 sends a vector of doubles to all other tasks for a variety of
> vector lengths. The following code works fine compiled with statically
> allocated memory, but fails when compiled with -DMALLOC  for message
> length in and around 16053. I find no errors running the code either way
> on a variety of other Linux/mpich setups.
>
> software levels: mpich/1.2.1..7/gm-1.5_Linux_beta3-2.4.13-ac2
>
> Is this a known issue with GM? Any help appreciated.

How much memory do you have on your nodes?  which kernel?
and when it fails do you see something in the kernel log?

Is it deterministic?  If this is the case, then can we have access to
the machine to try to reproduce this failure?

You could also try to recompile MPICH-GM to enable checksum,
and rerun your test program.  
  http://www.myri.com/scs/GM_FAQ.html#mpich25

I have taken your MPI program and run it with mpich-gm 1.2.1..7
and gm-1.5_Linux, and I cannot reproduce the error that you report.

So we would need to reproduce the error on site in order to investigate
further.

Thanks,
Susan

-- 
---------------------------------------------
Susan Blackford
Member of Technical Staff
Myricom, Inc.
---------------------------------------------
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.