Re: iSCSI/VI/TCP

[Costa Sapuntzakis <csapuntz@cisco.com>]

> Constantine Sapuntzakis writes:
> > A suitably generic RDMA scheme allows innovation in
                            ^^^^^^ = I should have used the word protocol here

> > protocol design, without worrying that the new protocol will be
    ^^^^ higher-layer protocol
 
> > hopelessly slow in legacy hardware.  It decreases the pressure to load
> > features into old, hardware-supported protocols instead of introducing
> > new ones. 
> 
> Here's where I must ask for a clarification. Are you saying that 

Thanks for pointing out the lack of clarity. I should have made the
layering in my argument more explicit. I am talking about innovation
in protocols at the iSCSI/NFS/CIFS layer.

Hypothetical world: The world starts building protocol-specific NICs,
for example, iSCSI NICs. There is a large installed base of iSCSI
NICs.  The NICs make data transfers on iSCSI run quickly while other
protocols go through the software stack.

In this world, let's say you want to create a new protocol/API that
competes with iSCSI/SCSI.  Well, to make it run quickly on the
installed base of iSCSI NICs, you would probably end up figuring out
how to run it on top of SCSI or iSCSI instead of creating a new
protocol. This puts pressure to dump features into iSCSI.

Admittedly, calling iSCSI "an old, hardware-support protocol" is a bit
premature.

-Costa