From panda at cse.ohio-state.edu  Mon Jun  5 23:23:40 2006
From: panda at cse.ohio-state.edu (Dhabaleswar Panda)
Date: Mon Jun  5 23:23:45 2006
Subject: [mvapich] Announcing the release of MVAPICH2 0.9.3 with
	multi-threading
In-Reply-To: <OF99F5A580.4299902F-ON86257180.006DFDCB-86257180.00723EEA@mck.us.ray.com>
	from "James D Ballew" at Jun 01, 2006 03:47:51 PM
Message-ID: <200606060323.k563Ne6I025470@xi.cse.ohio-state.edu>

Dear Jim, 

My apology for not able to reply back to your e-mail earlier. I have
been swampped in multiple deadlines ...

> Noticed this in the MVAPICH 0.9.7 User guide:
> 
> ??? (NEW) Support for Fault Tolerance
> ??? (NEW) Mem-to-mem reliable data transfer (detection of I/O bus error with
> 32bit
> CRC)
> ??? Additional fault tolerance support (such as checkpoint restart, automatic
> path
> migration (APM), etc.) will be introduced in successive releases
> 
> How far along are you on your support for Fault Tolerance in dual-rail
> networks.
> We are very interested in testing anything you have available and would be
> very interested in working with your group as a test site.

You must have come across the recent announcement of `full support
(both detection and retransmission) of mem-to-mem reliable data
transfer.

Since the current version of MVAPICH has separate single-rail and
multi-rail devices, most features are available on the single-rail
side because many people have been using this. Gradually, we are
migrating things to the multi-rail side.

In order to bring all features to single-rail and multi-rail side,
from the next release of MVAPICH2, we are unifying single-rail and
multi-rail designs (i.e., single-rail will be one instance of
multi-rail). After these, all others features will be available on
both sides (single-rail and multi-rail) on MVAPICH2 front.

Wrt fault-tolerant features like APM and Check-point restart, we have
these working in the lab as prototype forms. With more in-depth
testing, we plan to release these during the next 1-3 months.

Many thanks for offering your organization to work as a test-site. As
these features are completely available, we will definitely take your
help. In fact, for recent releases, we have been releasing RCs (after
some reasonable testing) through our SVN and announcing them on
mvapich-discuss mailing list. If somebody from Raytheon is on this
mailing list, they should be getting these e-mails and can try out the
early RCs.

Overall, as you know, we are in a university environment. Our progress
on adding new features completely depends on the demand from the IB
community and the available funding we have. We are always trying to
achieve a balance between these two. If Raytheon is interested in some
critical features and would like to extend some funding to us, we will
be happy to put more resource along these directions, work closely
with Raytheon and and accelerate on adding those features.

Please let us know your thoughts. 

Best Regards, 

DK

> Regards,
> Jim
>