Re: iSCSI Autosense Consensus, Connection next steps

["John Hufferd/San Jose/IBM" <hufferd@us.ibm.com>]

David,
I have seen some potential value in SCTP, however, it is far too young to
become a compelling proposal for Storage which will excite the industry and
the customers.

Also, I do not want SCTP to get in the way of the TCP/IP version of iSCSI
which has, so far, shown to have compelling interest to both the industry
and customers.

I know of numerous companies that are creating TCP/IP offload NICs etc. for
use in a lot of situations including iSCSI, and many that are creating
iSCSI-TCP/IP Device Drivers to use with their NICs.  I also know of a
number of companies that have created iSCSI Device Drivers that use OS
platform standard TCP/IP stacks.  Neither of these (HW or SW) will be able
to exploit iSCSI on a SCTP base very quickly.

Not only will a SCTP version require the HW vendors to change their HW NICs
to support SCTP (which they will be reluctant to do until they see a proven
business case for it, which is very hard to produce), but it will also
require the software only implementations to wait until the OS platform
supports SCTP (who knows when that will be, for any given OS).  Of course
the software vendors of SCTP/iSCSI could become their own implementer and
owner of an SCTP stack, in the various OSs, however since many of these
vendors were going to give their software iSCSI Device Drivers away, this
does not seem like good business.  All this tells me that an SCTP version
of iSCSI will cause the industry to push out the iSCSI for at least another
year or two.

Therefore, I think we need to focus on the TCP/IP version of iSCSI.  Upon
completion of that, if you feel there is sufficient value in a SCTP
version, then I can support IETF-ips creating that definition also.




.
.
.
John L. Hufferd