RE: [Tsvwg] Future of TCP and SCTP

Uri,

See comments interspersed-

On March 20, 2002 1:47 PM Uri Elzur (uri@broadcom.com) wrote:
>
> 1) In the ROI BOF yesterday, a statement has been made that "SCTP is a
> main stream Transport protocol" and that RDMA/DDP mapping to SCTP, will
> encourage application use of SCTP in the future.
>
> SCTP is not immune from IP packet drops and out-of-order reception, hence,
> like TCP a high speed flow-through NIC would either buffer out-of-order
> packets (expensive and not scalable at 10 Gbps and beyond) or place data
> out-of-order but indicate to ULP in order. (side comment: it seems to be a
> discussion on RDMA list, whether SCTP even provides the hints to the
> Receiver to enable placement of out-of-order data or buffering is required
> in any case)

When SCTP is used in the unordered mode, there is no buffering required in
the event of packet loss.  In this case it is immune from required buffering
in such events and would be very suitable for high speed data processing.
Hints are provided by the DDP shim that sits above SCTP as defined in the
specification written by Randall and myself.  Although this draft is about
to change to include a TLV style structure similar to SCTP for immediate
data needed by RDMA, this does provide requisite hints needed for any
reception order direct placement.

> 2) TCP = Protocol of the past ??? TSVWG agrees there is value in framing
> TCP as evident by admittance of draft-ietf-tsvwg-tcp-ulp-frame-00.
However,
> a single bit in the TCP header or a TCP option, that could have been used
> to make it simple, are "reserved for future use"

This TUF draft is not a standards track as I understand.  This draft also
has problems yet to be resolved.  In addition, it does not address improved
error handling.  Making new use of the TCP header bits may find such changes
highly detrimental to existing install base or simply not function.  If you
find a problem with SCTP, there is a much greater likelihood of it being
resolved.  The same is not true for TCP.  The IETF has strongly recommended
SCTP for this type of application.  DDP needs idempotent packets provided by
SCTP.

> What is the future direction for TCP? Isn't SCTP the future?

TCP has a future as TCP.  SCTP has a future as SCTP.  They have different
features where SCTP fulfills DDP needs where TCP supports legacy.  SCTP can
be adapted to handle TCP protocols however.  The SCTP-DDP draft helps with
partial write issues as example.

> Sorry if I missed something

If it were up to me, I would have booted this camel's nose from under the
tent.  Thanks for the soapbox.

Doug

> thx
>
> Uri

_______________________________________________
tsvwg mailing list
tsvwg@ietf.org
https://www1.ietf.org/mailman/listinfo/tsvwg