There is mention in the chelsio TOE thread that one reason you want TOE
nics is so that you can then do RDMAP on top of TCP on the adapter, and
get direct placement and kernel bypass...
----- Original Message -----
From: "Tom Tucker" <[EMAIL PROTECTED]>
To: "Roland Dreier" <[EMAIL PROTECTED]>; "Steve Wise"
<[EMAIL PROTECTED]>
Cc: <[email protected]>
Sent: Thursday, August 25, 2005 7:40 PM
Subject: RE: [openib-general] [PATCH][iWARP] Added provider CM verbs
andquery provider methods
Yes. I guess it's not obvious that the topics are related. Since
our RDMA transport runs on top of TCP and since the RDMAC verbs
spec requires that we have an LLP Handle (socket), we are dependent on
the resolution of this issue.
-----Original Message-----
From: Roland Dreier [mailto:[EMAIL PROTECTED]
Sent: Thursday, August 25, 2005 7:05 PM
To: Steve Wise
Cc: 'Roland Dreier'; Tom Tucker; [email protected]
Subject: Re: [openib-general] [PATCH][iWARP] Added provider
CM verbs andquery provider methods
Steve> See the current chelsio TOE thread on the netdev list.
Hmm, I just skimmed through this. It seems like a big flamefest about
TOE without any reference to RDMA.
I think someone from the iWARP world really needs to spend some time
explaining to the netdev community why we want to do connection setup
through the host stack so we can find a solution that everyone can
agree on. I don't want to bypass part of the community (and bypass
packet filtering too!) just because they'll "never" accept something.
- R.
_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
