[mpich-discuss] MPI I/O SLOW- possible conflicting withcommunication....code stuck for ever during IO

[Rajeev Thakur]

Can you send us a small test program that fails?


  _____  

From: mpich-discuss-bounces at mcs.anl.gov
[mailto:mpich-discuss-bounces at mcs.anl.gov] On Behalf Of hossam
Sent: Sunday, March 14, 2010 1:57 PM
To: mpich-discuss at mcs.anl.gov
Subject: [mpich-discuss] MPI I/O SLOW- possible conflicting
withcommunication....code stuck for ever during IO


Hi,

I think i have a well defined problem i wonder if someone can give an
advice.

I am running an MPI code (MPICH2) on a linux cluster. 

When i run the code with no IO inside the main runtime loop. It works
perfectly fine on any number of processors and nodes. When i start
dumping data during the loop
the code get stuck for ever after the second or third file. 

Everything works perfectly fine on a single processor though.

Now, to make things clearer: When i use ROMIO (MPI IO) it got stuck
during opining the file (MPI_FILE_OPEN). 

However, i tried regular open(....) at the root processor and then
reducing the data to the root processor and write there. The code got
stuck at the MPI_REDUCE call.

If i eliminated all the processors communication and just do dump IO
(both ways), it runs fine but it slows down significantly when dumping
the data.

I tried on three different clusters(all linux) and i got the same
problem.

I don't want say what i think is the problem, i don't really know, but
if the experts here can help i will appreciate.

 Thanks


On Sun, Mar 14, 2010 at 12:50 PM, hossam <hossam.elasrag at gmail.com>
wrote:


Hi,

This is my first post to the mpi-forum. I have a runtime problem related
to MPI using MPICH2. I wonder if this is the correct forum to post my
problem.
Please let me know so i can proceed forward with my problem details.

Thanks










-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20100315/a87c46a1/attachment.htm>