[mpich2-dev] A Question about romio-hints.

["Kwangho CHA"]

Rob,

Thanks for your advice.

When I sent the email, my MPI program failed to open a file ( It seemed there were some problems in mapping new aggregators.). 

However, as you recommend, I installed new version of mpich2 (1.1b1) and it made the problem go away.

Thank you again.

Kwangho CHA.



----- Original Message -----
From: "Rob Latham" <robl at mcs.anl.gov>,
To: <mpich2-dev at mcs.anl.gov>,
Date: 2009-04-28 06:00:51
Subject: Re: [mpich2-dev] A Question about romio-hints.

On Sun, Apr 26, 2009 at 08:30:59PM +0900, "Kwangho CHA" wrote:
> Dear All.
> 
> I've interested in the effect of the aggregators in collective I/O
> and tried to change aggregator nodes using "/etc/romio-hints" file.

Great! I'm happy to see someone else using that feature.

> However, my MPI program was terminated abnormally, when I described
> "/etc/romio-hints", as follow;
> 
>  cb_config_list compute-0-0.local,compute-0-1.local,compute-0-2.local,compute-0-3.local
> 
> Instead of this line, I also described "cb_config_list *:1", but it
> showed the same result.
> 
> When I comment out the line of the romio-hints file, my MPI program
> works well.
> 
> What should I check first?

When you say 'terminated abnormally', what happens?  Does it dump
core?  Does it exhibit incorrect behavior?  How sophisticated a test
program do I need to reproduce this?   Do any of the tests in
romio/test show the problem?

> My system consists of mpich2-1.0.8p1 and pvfs-2.8.1. 

I may have fixed this in SVN.  Can you either try with the 1.1b1
version or apply the attached patch to your tree?

Thanks
==rob


-- 
Rob Latham
Mathematics and Computer Science Division
Argonne National Lab, IL USA