To the best of my knowlege, I'm the only guy working on MPI extensions to
DX, and I've been working on it for quite a while - since about the time
that DX was open-sourced, in the '98-'99 time frame. I took a false start,
which I generally called DDX, in which I attempted to spread the OpenDX
data model across multiple memory spaces by substituting distributed
pointers - (node #, phys address) for the simple pointers used to link
data-mode objects together in OpenDX. That lead to lots of problems, not
the least of which was the vast number of source-code changes all through
the OpenDX executive, leading me to be very doubtful about ever getting it
reasonably stable for real use. I think that the current approach is a
vastly superior one, allowing the operations on distributed data to be done
at a very high level, such as the use of parallel macros - networks of
modules, running in parallel on the data processing nodes on local data.
DDX, in contrast, operates on distributed data at a very low level, where
any simple data operation - such as a deletion, may require the suspension
of exection on the caller while a request to perform the operation is
transmitted to the node where the data resides, the op is performed, and a
acknowlegement shipped back to the caller. It also requires virtually no
source-code changes to the OpenDX source base, since its implemented
entirely on top of OpenDX as a set of custom modules that implement the
distributed representation and roll in the necessary communications code.
If you are interested in OpenDX/MPI, I'd be happy to give you access to the
source code.
Greg
Aaron Carass
<[EMAIL PROTECTED]> To:
[email protected]
Sent by: cc:
[EMAIL PROTECTED] Subject: Re: [opendx-general]
OpenDX with MPI
son.ibm.com
06/10/2002 03:21 PM
Please respond to
opendx-general
>>>>
>> Status is that its working fairly well but very poorly documented. I
can
>> distribute it to academic-associated people, but am still trying to
figure
>> out how to get IBM to OK a full open-source release.
>>>>
I presume I can qualify as an academic-associated person ?
Given that, how can I get my grubby hands on this package.
Also this package is not what I originally had in mind, I wanted to get
dxmpi which was under development back in 99 or 2000. So if anyone has
information about this, I would be grateful.
Aaron
>
>
> I answered this one a couple weeks ago.
>
>
http://opendx.watson.ibm.com/dx/mailArchives/mails.html/opendx-users.0205/msg00062.html
>
> Greg
>
>
>
>
--
Aaron Carass
Image Analysis and Communications Lab.
Johns Hopkins Univ.
