[openib-general] putting in dead wood for DAPL and similarabomination

Clem Cole clemc at ammasso.com
Thu Mar 3 07:24:35 PST 2005


Thanks for the kind words Tom.

Indeed, Ammasso is fully open source and are thrilled at the
idea of getting DAPL into the mainline.  We went GA with our 1.2 release
recently and have had our AMSO100 hardware and associated iWARP software
in the hands of about 60 different sites - both HPTC and commerical.

Feel free to download the code and have a look.  As Tom says it has been
tested
on kernels as far back as those for RH 7.3 and modern as kernel.org's
2.6.10.
We have tested on both x86 and x86-64.  

As Tom says, we too (as well as our customers) are statisfied
users of the DAPL interface.  DAPL certainly continues to show that it
works
pretty well for us and is easier to use than the low level QP verbs
layer.
Recently an ISV moved a hunk of code from another interface (we do not
know
which) to our DAPL implementation in about 2.5 weeks.  

For the record, our kDAPL and uDAPL are dervived from the DAT reference
code.
Besides writing the Ammasso specific provider code, since IB != iWARP
we did have to make some small changes to common code to handle iWARP
specific difference. We are working with Tom, Arkady and the rest of the
DAPL community to get the iWARP changes back into the base to help
ensure
that the DAPL interface is not considerred just an IB thing or that
people 
come to the incorrect conclusion that IB verbs are ``good enough.''

For whatever its worth, we are actively working on not only being
DAPL/iWARP >>compliant<<
(working with UNH etc) but also >>compatible<< - going to plugfests and
working with
actual ISVs that have written linux code that rely on DAPL/iWARP -- i.e.
not only do we
pass the full uDAPL/kDAPL test suite, we have been working with a number
of different large
commerical vendors (who's source code we have never seen) to get their
tests
as well as their >>applications<< which were designed to run over DAPL.
Since
these codes had been previously only tested on IB (i.e we are the first
shipping iWARP provider), we feel pretty good that our DAPL works as
expected.

Note: our plan is to increase the number of applications that use the
code as quickly as possible; but under we are small start up and are
band limited by the number of ISV we can work directly at one time.

If you have specific questions, feel free to take them off line to me.

Clem Cole
Dist. Eng

PS If you are interested in getting your hands on hardware,
drop me a line and I'll make connection to our sales guys.

-----Original Message-----
From: Talpey, Thomas [mailto:Thomas.Talpey at netapp.com]
Sent: Thursday, March 03, 2005 9:47 AM
To: openib-general at openib.org
Cc: Clem Cole
Subject: RE: [openib-general] putting in dead wood for DAPL and
similarabomination


At 02:22 PM 3/2/2005, Woodruff, Robert J wrote:
>I think the point is that only one of those interconnects (IB) is 
>in the kernel, the rest are proprietary. Do any of the other RDMA
>interconnect vendors plan to submit their code for inclusion into Linux

>in the near future ?

Yes - take a look at <http://www.ammasso.com/support.html> where
you can freely download their complete stack, including drivers for
their
iWARP NIC, plus MPI and DAPL API libraries. It runs on 2.6.10 and many
versions back (including 2.4.x).

I'll let Clem speak for his plans to submit it, however. I'm just a
satisfied
user.

Tom.




More information about the general mailing list