Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |  

This web mail archive is frozen.

This page is part of a frozen web archive of this mailing list.

You can still navigate around this archive, but know that no new mails have been added to it since July of 2016.

Click here to be taken to the new web archives of this list; it includes all the mails that are in this frozen archive plus all new mails that have been sent to the list since it was migrated to the new archives.

Subject: Re: [OMPI users] nonblocking MPI_File_iwrite() does block?
From: Rob Latham (robl_at_[hidden])
Date: 2010-01-06 17:35:55


On Mon, Nov 23, 2009 at 01:32:24PM -0700, Barrett, Brian W wrote:
> On 11/23/09 8:42 AM, "Rob Latham" <robl_at_[hidden]> wrote:
>
> > Is it OK to mention MPICH2 on this list? I did prototype some MPI
> > extensions that allowed ROMIO to do true async I/O (at least as far
> > as the underlying operating system supports it). If you really need
> > to experiment with async I/O, I'd love to hear your experiences.
>
> Rob -
>
> Funny you should mention that code. I was looking into a compile error in
> OMPI on a strange platform last week (while travelling to SC, of course) and
> was realizing that I had no idea what I was doing when I integrated ROMIO
> into Open MPI many years ago, did it horribly, and want to clean it up.
>
> Anyway, you're talking about all the MPIX_Grequest interface used in ROMIO,
> right?

exactly

> Do you have documentation of that API anywhere (even a header file
> with some comments)?

There's the EuroPVM/MPI paper:
http://www.mcs.anl.gov/~robl/papers/latham_grequest-enhance.pdf

> I'm going to try to clean up how we integrate into ROMIO during the
> holiday break, and it might make sense to implement the extended API
> so that we can give some async I/O behavior as well. From looking
> at the ROMIO usage, I think OMPI has all the internal pieces to put
> together the API pretty quickly, so it might be an option. Any
> thoughts you have on the subject would be greatly appreciated.

I'm sorry I did not respond to this message earlier. I only check my
openmpi folder every few weeks.

When we implemented this in MPICH2 we added additional checks in our
{test,wait} {,any,all,some} routines and invoked the progress function
at those times. It resulted in some less-than-efficient code.

==rob

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