[tcpm] "David's proposal" for cc class terms (was RE: [tsvwg] L4S status tracking)

["Black, David" <[e-mail]>]

In more detail than I explained earlier: in this forum

We are dealing with two classes of congestion controls.  For lack of better terms the following class names are based on what the transport protocol throughput is proportional to where 'p' is the loss and/or congestion marking probability:
               - 1/sqrt(p)-class congestion controls: Includes most existing TCP congestion control algorithms, e.g., NewReno, CUBIC.
               - 1/p-class congestion controls: Includes DCTCP congestion control.
Keep in mind that p is a probability that is usually << 1 when expressed as a decimal, e.g., p=0.01 represents a 1% loss/marking rate.

Thanks, --David

From: tsvwg <[e-mail]> On Behalf Of Bob Briscoe
Sent: Wednesday, November 13, 2019 5:00 AM
To: Scharf, Michael; Wesley Eddy; Rodney W. Grimes
Cc: [e-mail]; [e-mail]
Subject: Re: [tsvwg] [tcpm] L4S status tracking


[EXTERNAL EMAIL]
Michael,

I see the problem. Nearly all the occurrences of 'Classic TCP' are in the appendix that was written v early on, when the focus was only TCP. That text was not written by me, and I haven't got round to editing it for consistency. However, there are 4 occurrences of "'Classic' TCP" in the abstract and intro, which I will also fix.

Can you say which "David's proposal" you mean please?


Bob
On 11/11/2019 23:10, Scharf, Michael wrote:
OK, so we seem to make some progress at least regarding the term 'classic' TCP. As mentioned before, I personally would find the term 'Classic' QUIC very confusing and you cannot avoid that if you talk about 'Classic' TCP, IMHO. So, it seems better to avoid that terminology for all transport protocols. For the record, in draft-ietf-tsvwg-ecn-l4s-id-08 the combination of the terms 'classic' and 'TCP' back-to-back is used 11 times as far as I can see.

Regarding congestion control, I'd like to emphasize that David's proposal would also work well for me, i.e., characterizing the congestion control by functional properties. This avoids all issues regarding whether something will supersede something else (including the issue that EXP is just an experiment that won't update or replace any PS document).

Michael

From: Bob Briscoe <[e-mail]><mailto:[e-mail]>
Sent: Monday, November 11, 2019 4:31 PM
To: Scharf, Michael <[e-mail]><mailto:[e-mail]>; Wesley Eddy <[e-mail]><mailto:[e-mail]>; Rodney W. Grimes <[e-mail]><mailto:[e-mail]>
Cc: [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>
Subject: Re: [tcpm] [tsvwg] L4S status tracking

Michael,
On 07/11/2019 08:11, Scharf, Michael wrote:
... and just to show that _my_ concern would be trivial to address by small editorial changes:

For the abstract of draft-ietf-tsvwg-ecn-l4s-id, the wording "It gives an incremental migration path so that normal TCP traffic classified in the 'Classic' service of L4S will be no worse off" would already work for me.  It just takes a small editorial change to make me happy. This can't be so hard.

Personally, I wonder if "classic" is indeed the best name for a service class (and, e.g., "normal" sounds better to me), but in the context of a service class, "classic" could actually work for me, if TSVWG really wants that name with strong consensus. I don't care how about names for DiffSe^D^D^D^D^D^D traffic classifiers, traffic policers/shapers, AQM schemes, or whatever else is done in the fast path of a router to implement low latency service.

I only object to specific terminology such as 'Classic' TCP or 'Classic' congestion control because I don't think that 'classic' is a proper characterization for TCPM standards or TCP/IP stack behavior not aligned with whatever L4S believes the bright future shall be.
I thought I had already agreed to remove 'TCP' from all instances of 'Classic TCP traffic' etc, and that I hadn't intended to include 'TCP' in the first place. Because it's the congestion controls that are relevant in L4S drafts, not the protocols.



Bob



Whether such small rewording address fully the concerns from others may be a different question.

Michael



From: Wesley Eddy <[e-mail]><mailto:[e-mail]>
Sent: Wednesday, November 6, 2019 8:54 PM
To: Scharf, Michael <[e-mail]><mailto:[e-mail]>; Bob Briscoe <[e-mail]><mailto:[e-mail]>; Rodney W. Grimes <[e-mail]><mailto:[e-mail]>
Cc: [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>
Subject: Re: [tcpm] [tsvwg] L4S status tracking

On 11/6/2019 1:57 PM, Scharf, Michael wrote:
Bob,

>From draft-ietf-tsvwg-ecn-l4s-id-08: "It gives an incremental migration path so that existing 'Classic' TCP traffic will be no worse off"

You are proposing an experiment. Not more than that. I will be fine with the term "Classic" for TCP and TCPM-specified congestion control when more than 50% of Internet traffic uses that new technology.

Until this happens, I insist that the word "Classic" must be removed in all context of TCP and congestion control (as far as it is owned by TCPM), including the reference above. BTW, "normal" as suggested would also work for me. So, you have plenty of options for other terms.




If Dave+Michael's suggestion of replacing "classic" with "normal" is agreable to others, this seems like a good way forward to me.  It should be easy enough to explain in other SDOs that classic and normal mean the same thing, if this is a real issue.

(FWIW, I've never had a problem myself with "classic", nor read any negative connotations to it.  However, for the sake of working group progress, I think we just need to pick something that seems the least terrible and agree to move on.)






--

________________________________________________________________

Bob Briscoe                               http://bobbriscoe.net/



--

________________________________________________________________

Bob Briscoe                               http://bobbriscoe.net/