RE: Last call comments on SCTP RDMA/DDP Adaptation draft

"Caitlin Bestler" <[email protected]> Wed, 6 Apr 2005 10:00:43 -0700
Newsgroups gmane.ietf.rddp
Message-ID <[email protected]>
Hemal,
 
Thanks for a thorough proofreading.
 
Most suggestions are noted, and I will update accordingly
after all changes are collected.
 
There are a few exceptions, noted in-line below.
 
> 4.       Section 2.2.1: This section does not say how DDP-SSN
>          is incremented. Suggest adding the following text
>          "DDP-SSN is increased by one (modulo 65536) with each
>          DDP message submitted to the SCTP layer."

It is once per SCTP Data Chunk, which would be once per DDP Segment.
That is clear in the definitons section, but I agree that repeating
that definition in section 2.2.1 would improve the document.

> 5.       Section 2.1: Need to add a requirement that DDP-SSN for
>          the DDP Segments of a DDP message should be same. Suggest
>          adding the text: "The DDP-SSN field of the Payload Data
>          Chunks of the DDP segments of a DDP Message MUST be the
same."

The DDP-SSN numbers DDP Segments/SCTP Data Chunks, not messages.

One of its purposes is to allow easy determination when all prior
segments have been received for normal DDP ordering rules. When the
DDP Segment is untagged and has an L bit, it generates a completion.
But only after all prior segments have been received. The difference
with the SCTP adaptation versus MPA is that the SCTP adaptation uses
a 16 bit sequence number with 1 tick per segment rather than a 32 bit
TCP sequence number that has n ticks per segment.

> 6.       Section 2.2.1: This section does not mention whether DDP-SSN
>          is passed to DDP layer or not. For delivery, it is important.
>          Suggest adding the following text "The adaptation layer
passes
>          DDP-SSN to the DDP layer for the in-order delivery of DDP
Messages." 
	 

The DDP specification already requires that "ordering information" be
supplied
from the LLP, so no explicit statement is really necessary. But if one
were
added it should be more along the lines of "The adaptation layer passes
DDP-SSN
to the DDP layer as the required LLP ordering information to allow the
DDP layer
to sequence DDP segments properly." 

This avoids suggesting that the DDP Segments are delivered to the DDP
layer in order.

> 7.       Section 2.2.1: Who maintains DDP-SSN, adaptation layer or
>          DDP layer? Please clarify it in this section. 

	 

I'll think about the best clarification. The DDP-SSN is fully equivalent
of
the TCP sequence number, it is the property of the LLP. It is also
basically
equivalent to what the SCTP Stream Sequence Number would have been had
ordered
delivery been used. Unfortunately using ordered chunks would have
undesirable
effects, such as causing many SCTP stacks to hold Data Chunks until they
were
in order. Use of unordered data chunks reduces the complexity of the
required
SCTP functionality and guarantees the SCTP stack will forward the data
chunks
to the adaptation layer as promptly as possible.