RE: SCTP [WAS: Re: Towards Consensus on TCP connections]

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

Costa,

I understand the desire for a conservative approach.  SCTP is based on the
mature TCP/IP algorithms so SCTP features added are squarely aimed at
providing an object rather than a byte stream, a feature SCSI needs and yet
ignored due to these slight structural changes.  Too new?  The iSCSI layer
took the opposite approach.  Rather than adopting existing serial structures
in prominent use in high-end SAN facilities, these structures were discarded
in toto.  This will ensure tools and drivers will require extensive
modification, in addition to existing equipment made incompatible.  Too old?
Can you explain this conservative/non-conservative choice a little more
fully and provide a bit of detail or history on the harm caused by adopting
SCTP and FCP structures?

It would not be impossible or impractical for FC over IP/SCTP to be a
solution for next generation products.  It would not be difficult to make
such a specification or to implement.  Tools, drivers, and equipment
investments remain and system developers have fewer exceptions in what is
already difficult.  FC has management structures that could implement the
bridge flow adjustments as well as other tricks in buffer management.

Doug

> -----Original Message-----
> From: owner-ips@ece.cmu.edu [mailto:owner-ips@ece.cmu.edu]On Behalf Of
> csapuntz@cisco.com
> Sent: Friday, August 18, 2000 9:40 AM
> To: ips@ece.cmu.edu
> Subject: SCTP [WAS: Re: Towards Consensus on TCP connections]
>
>
>
> SCTP is not stable currently and there is very little experience with
> it. This does not seem like the best route for a fairly conservative
> field like storage. A later organization, once SCTP is mature, can
> once more reconsider it.
>
> -Costa
>