[Rdma-developers] Re: [openib-general] OpenIB and OpenRDMA: Convergence on common RDMA APIs and ULPs for Linux

[Caitlin Bestler]

> -----Original Message-----
> From: Grant Grundler [mailto:iod00d at hp.com] 
> Sent: Friday, May 27, 2005 10:10 AM
> To: Caitlin Bestler
> Cc: Sukanta ganguly; rdma-developers at lists.sourceforge.net; 
> openib-general at openib.org
> Subject: Re: [Rdma-developers] Re: [openib-general] OpenIB 
> and OpenRDMA: Convergence on common RDMA APIs and ULPs for Linux
> 
> On Fri, May 27, 2005 at 09:58:44AM -0700, Caitlin Bestler wrote:
> > > I understand most of what you wrote but am still missing one bit:
> > > How is the RNIC told what the peer IP is it should 
> communicate with?
> > 
> > The TCP Connection is already established before the RNIC takes 
> > control of the connection. Therefore the entire TCP state 
> is already 
> > established (source/dest IP Address/Port, negotiated options, EMSS, 
> > local options, etc.).
> 
> Would "shares the connection" be a better description?
> 

No, it does not "share the connection". After conversion the 
RDMA stack is in full control of the TCP layer. The host cannot
modify anything about the TCP layer, or use its services.
For example the host cannot enable Nagle on the TCP connection
after it has been turned over to the RNIC. It definetly cannot
send or receive streaming data. Because of the nature of the
MPA protocol, a TCP connection that has been enabled for MPA
can *never* revert to non-MPA mode. It can only be used in MPA
mode or terminated.

It's the L2 and L3 layer data that is shared (or synchronized)
between the host and the RNIC.

> I think I get the concepts at least now...this follows what I 
> was suggesting earlier about having an RNIC support both NAPI 
> (New API for NICs) (code under drivers/net) and openib.org 
> Gen2 stack (code in drivers/infiniband).
> 
> thanks,
> grant
>