RE: New List: rdma@cisco.com: to discuss RDMA

["Douglas Otis" <dotis@sanlight.net>]

Steve,

There should be two generic automatic delivery schemes.  One that allows a
remote agent to determine what data is to be delivered and where.  This
would be VI for such things like DAFS.  With such a scheme, specifics of
where to put data is sent with the data after negotiations of intimate
details with highly trusted systems.  The other is SCSI where location is
understood by reference and only the offset is included, as such,
scatter/gather is done locally in a far safer manner.  FCP structures are
stable for such use.

As far as either hardware or software, SCTP would allow both without
interference.  There would be no hard or easy way, but simply one way.  As
hardware became available, these features could be added without any noticed
change to the data structures or placement.  The concepts of RDMA is
essentially a restructuring of TCP to aid hardware assistance.  Such is not
required for SCTP nor would TCP safely allow out of sequence processing
without restructuring.  SCTP can be implemented without kernel changes
whereas RDMA require extensive changes to the TCP  kernel and API.  A clean
and clear path is SCTP for both hardware, software, kernel and non-kernel
implementations.  In addition, to achieve stability with SCSI structures,
retain FCP definitions.  These FCP structures have been implemented to
facilitate such hardware acceleration.

Doug

> -----Original Message-----
> From: owner-ips@ece.cmu.edu [mailto:owner-ips@ece.cmu.edu]On Behalf Of
> Stephen Bailey
> Sent: Thursday, September 28, 2000 2:23 PM
> To: ips@ece.cmu.edu
> Subject: Re: New List: rdma@cisco.com: to discuss RDMA
>
>
> > However, I would support structuring the spec so that an RDMA
> > transport mechanism could be used underneath (I guess that's
> > motherhood).
>
> Not necessarily.
>
> You have to ask the implementors (particularly the hardware
> implementors), what sort of optional RDMA proposal they'd be happy
> with.  My answer is none.  It's mandatory or not at all.
>
> The reason for using RDMA is to make the implementation of iSCSI
> easier in hardware.  If there are implementations which do not support
> the RDMA protocol, then your hardware implementation will have to
> support both the `easy path' (using RDMA) and the `hard path' (no
> RDMA).  If you have to implement the hard path anyway, there's no
> point in implementing the easy path.
>
> The argument that you could make the hard path infrequent and
> implement it in software doesn't wash in this case.  It just takes one
> implementation that doesn't do RDMA to slow your system to a crawl,
> and the competitor who only implemented the non-RDMA path makes you
> look like a fool.
>
> Fundamentally, RDMA has to be either adopted or punted.  Of course,
> I'm happy to have somebody prove this statement wrong.
>
> Steph
>