Re: [core] WG Last Call on draft-ietf-core-new-block (No-Response)

["Christian M. Amsüss" <[e-mail]>]

Hello Med,

On Wed, Feb 03, 2021 at 08:09:40AM +0000, [e-mail] wrote:
> As a follow-up on this issue, we finally went with an approach that
> does not require No-Response. We do have the following to avoid extra
> delays, e.g., 

The need to indicate when a client wants confirmation has been avoided
by having timeouts on both sides, and relying more on MAX_PAYLOADS being
aligned.

For my taste, that makes unnecessary sacrifices in robustness over the
simple two-byte NoResponse:0 that'd go with every MAX_PAYLOADS'th
Q-Block2 from the client (and would completey avoid the risk of stalling
in a lossless situation).

(For the other direction, corective heuristics can easily be applied).

But given they are explicitly configured in the DOTS case where Q-Block
is used, so -- well. Food for a block-bis (or generalized additions that
allow block to be used that way).


That aside, in the interest of readers who try to understand the
ecosystem and the trade-offs involved, I think that at least a reference
to it in a statement like "was not chosen because of X" would be
helpful. The "Note that similar performance benefits" section would lend
itself to that. Text suggestion:

  [messages are provided in Appendix A.] Similarly, The No-Response option
  can be applied to Block1 requests, but offers no corresponding facility
  for Block2 responses.

BR
c

-- 
To use raw power is to make yourself infinitely vulnerable to greater powers.
  -- Bene Gesserit axiom