[ofa-general] [RFC] dropping RDS over TCP support

[Jon Mason]

On Tue, Sep 02, 2008 at 01:50:39PM -0700, Andy Grover wrote:
> Jon Mason wrote:
> > On Tue, Sep 02, 2008 at 01:04:19PM -0700, Andy Grover wrote:
> >> We've been discussing dropping RDS's support for using TCP as a
> >> transport, and just focusing on RDS as a IB and iWARP-focused protocol.
> 
> > I found it useful for early development of RDS iWARP support. 
> 
> Hmm! Were you actually running it or just using it as a reference?

I was running it to see if there were any issues with iWARP and IB
co-existing (as I was developing a stand-alone iWARP RDS method at the
time).  It does work, contrary to the documentation.  

My only reason to keep it in would be for those developers who do not
have access to IB/iWARP hardware.  There may be people who would want to
sue it for something beyond is current usage bad this would lower the
bar of entry....but I can't imagine who they are or why they would want
to.

> > While I
> > believe that mainline inclusion is much more important, I don't see any
> > harm in keeping it around.  Is there a thread on lkml listing it as an
> > issue for mainline inclusion?
> 
> We haven't brought up rds inclusion on lkml yet, but I think positioning
> RDS as an IB protocol is a good thing. RDS may be a better fit in
> drivers/infiniband/ulp/rds, rather than net/rds, for example.

Removing the TCP module would remove the need for it to be in the net/
dir (as it would then be only IB/iWARP).  It will also remove the need
to have it go through the netdev mailing list..which may or may not be a
good thing.

> Removing unused code is also always a good thing.
> 
> Lastly, we're looking to extend rds with more ib-centric features in the
> future, so it would be a development burden to tunnel that over the TCP
> transport...and that starts to sound a lot like reinventing iwarp ;-)

I have no issues with removing it.  I simply wanted to make sure that is
is not being used.

Thanks,
Jon

> Regards -- Andy