Re: [ippm] Comments on draft-ietf-ippm-explicit-flow-measurements-00

["Lubashev, Igor" <[e-mail]>]

Ike,

Thank you very much for reading the draft and your very useful comments. It is especially important to understand where readers may get misled.  We are working on addressing your comments and suggestions for the next version.

As for the organization of the document, especially separating the sections describing individual bits and combinations of bits, we can certainly do better.  But I wonder if it is wise to separate the discussion of measurements facilitated by R bit from the discussion of measurements facilitated by R+Q bits, because R bit requires Q bit for its implementation.  The L bit allows for some useful measurements by itself, but its most valuable measurements are realized in combination with Q bit.

So the only “clean” alternatives I see are:

  1.  Discussing measurements facilitated by L+Q and R+Q bits as sub-sections of L and R bit sections, or
  2.  Leaving only the discussion of the mechanics of bit generation and observation in the bit sections and moving all the measurement discussions to a separate section (with “T bit measurements”, “Q bit measurements” , “L bit measurements”, “L+Q bit measurements”, “R+Q bit measurements” as subsections).

I am worried that alternative B may be more confusing for the reader, since it would keep the details of the algorithms far from the algorithms’ indented purpose.  But I welcome input here.

Many thanks!


  *   Igor


From: Ike Kunze <[e-mail]>
Sent: Friday, March 18, 2022 11:49 AM
To: [e-mail]
Subject: [ippm] Comments on draft-ietf-ippm-explicit-flow-measurements-00

Hi all,

in my mail supporting adoption of the efm draft, I mentioned that "the descriptions of the approaches could be a bit more precise”.
In the past weeks, I have discussed these aspects with Mauro, Massimo, and Fabio and would like to quickly summarize my main “concerns” here so that (hopefully) everyone can profit.

R-Bit:
To me, the exact behavior when updating the R block length if a new Q block has been completed during the R block transmission is not entirely clear.
While the text on this is technically correct, I believe that it could be beneficial to explicitly describe the behavior, e.g., when the new R block length is shorter than the previous one.

T-Bit:
I believe that the description of the T-Bit could profit from a more detailed presentation of two aspects:

First, I misinterpreted that the generation token counter is reset each time a new generation phase starts.
This then led to several additional false inferences.
Consequently, I think that the description of the concrete behavior of the generation token counter could be more precise.

Second, the concrete choice of “2 spin bit periods” for the T-Bit phases (in connection with the misinterpretation of the generation token counter) again led to
several false inferences on my part.
Thus, I think that some discussion on why this phase length is chosen might be beneficial to avoid such misinformed reasoning.

Additionally, the T-Bit algorithm is currently presented once at a semi-high-level at the start of Section 4.1 and then again in Section 4.1.2.
This made me jump between the sections several times as it is not obvious which part contains which information.
My suggestion would be to either (i) merge the two parts or to (ii) make the description at the start of 4.1 more general so that it is clear that it is only intended as a rough overview and not as potentially containing critical implementation aspects.
That could help so that all information is found in one place and one place only.

General comment:
In addition to these aspects directly concerning two of the mechanisms, I also have a more high-level proposal regarding the structure of the loss bit section.
In Section 4, the combination of L+Q comes as Section 4.4. after the T, Q and L Bits have been presented.
The R-Bit is then presented in Sec 4.5 and the combination of R+Q in subsection of 4.5.1.
I think it would be cleaner if all of the loss bits would be presented first.
The different possible combinations could then be presented in a joint Section 4.5.
In my opinion, this would make it easier to fully grasp the different possible combinations and, especially, allow for a sensible comparison between the possibilities.

Hope this helps!
Cheers,
Ike