Re: [babel] Information model: fast-convergence

[Donald Eastlake <[e-mail]>]

Hi,

On Mon, Jul 27, 2020 at 12:00 PM Juliusz Chroboczek <[e-mail]> wrote:
>
> Dear all,
>
> Barbara proposed a switch called (if memory serves) "fast-convergence"
> that is essentially a macro that changes the values of a number of
> parameters to values that make convergence faster at the cost of higher
> overhead. ...
>
> It is my opinion that this switch should not be added.

Speaking as a WG member, I agree.

> First of all, the values given in Appendix B of 6126bis are just examples;
> the protocol is robust enough to function with a wide range of parameters,
> and I expect implementations to use different sets of parameters in
> different kinds of networks.  (The only reason they appear as "recommended
> defaults" is because some IESG members appear to be unable or unwilling to
> accept the notion of a robust algorithm.

I do not think it has anything to do with non-acceptance of algorithm
robustness. I believe their thought process is more or less as
follows: If a parameter makes no difference, then it shouldn't be
called out at all or should just have a fixed value. If a parameter
does make a difference then, while this might not apply to
informational or experimental or research documents, for a protocol
standard there should be a recommended value or a method for
determining a recommended value.

As an example I recently saw, look at Section 14 (top of page 16) in
https://tools.ietf.org/html/draft-ietf-6tisch-msf-16 (I read this
draft because I was asked to review it as a member of the IETF
Security Directorate.) That document calls them "constants" rather
than "parameters" but it is the same thing and the draft has a table
of their RECOMMENDED values. It is what people want and expect in a
protocol standard, at least in the IETF and in IEEE 802 where I have
the most experience.

>                                       Please ignore this recommendation,
> and experiment with the values most suitable in your particular network.

A recommended value is just that: only a recommendation.
Implementations are free to use other values and will still be
conformant to the standard.

Thanks,
Donald
=============================
 Donald E. Eastlake 3rd   [phone] (cell)
 2386 Panoramic Circle, Apopka, FL 33896 USA
 [e-mail]

> ...
>
> -- Juliusz